Electronics - High frequency metal vapour arc lamp power supply

This high-frequency ballast for a metal vapour arc-light source works in a similar way to a conventional 50Hz mains power source and choke but working at high frequency regulated AC current source instead. The conventional ballast is different because it also has a fluorescent tube type starter or the ballast is a step up leakage transformer type.

Summary

I have given myself a design objective to improve the power supply's efficiently and to update the parts originally used if necessary.  The Arc-lamp power supply design involves a lot of iterations of various models.  The requirement is unusual in that starting up voltage and power circulating in the power supply (not the lamp) is very high and may be >5KW briefly although the power consumed is much less than this.  After starting and warming up the unit needs to run efficiently at 10-60W. 

In this case, it was useful to get a functioning though inefficient, robust design to work in order to prove the design idea and the understanding of the lamp.  Many of the design iterations are not included but those that are included are not ideal at that stage.  By using new transistor type and changing the snubber R+C to just a capacitor, similar to resonant mode an SMP the transistor losses simulated become much lower.  This particular project is unusually complex to simulate therefore it will be necessary to prototype the revised design but it is very likely a heat-sink for the transistor can be substituted by an area of copper on the PCB.  In any case, a PTC thermistor is required under the high side transistor to shut-off the power supply if the lamp fails and this will require some trial and error work.

Recent changes; - it is important to keep a change and a To-do's lists although it is not necessary for this blog.

• 20-12-2018 This project has been submitted to contest; Powering the future with Infineon; https://www.hackster.io/contests/infineon-coolmos 
• 21-06-2019 Spelling and grammar check.
• 17-08-2019 Minor corrections.

Conventional mains frequency ballast;

The metal vapour arc lamp's arc is struck with a high voltage of 900V peak (GE fluorescent tube starter data-sheet) and in the high-frequency lamp, a limit was set at 1.5KV peak which works reliably.  The noble gasses ionise and begin conducting current. In the case of a Sodium (Na), lamp voltage quickly drops to 500V then operates at 100V for a short time, heating up the metal until it becomes a vapour within the lamp.  During this short period, it is likely that the lamp unit will behave like a spark-gap transmitter.  The lamp voltage then falls to about 15V at 1A as the metal vapour conducts current and the lamp appears to have a resistive characteristic.

The mains lamp unit appears to re-strikes each mains cycle 2x50Hz so could cause emissions all the time it is running that can be observed with an oscilloscope or a spectrum analyser.  Also, there is some blue and flickering light.

I found this video on YouTube of a sodium vapour arc lamp starting.  You will see 
some blue and other colours around the electrodes plus flicker and fluctuation in this 

conventional mains power ballast when it is fully warmed up and running normally. 

When warmed up and running normally that blue and other colours plus flicker and fluctuation are much less apparent in the high-frequency metal vapour-arc-lamp power supply.  The circuit below is different from a modern electronic ballast lighting in that the lamp's current is constant and regulated, that is, the current waveform is not modulated by rectified AC mains voltage.

Starting with guesswork and developing a sound theory leads to a robust solution;
The starting and running characteristics were all discovered by observation and experimentation, as described above, but with a range of different metal lamps.  The data-sheet on the lamps has very little information - it took me a while to find it.

This project went through a lot of iterations and experimenting that concluded with a simple looking solution that was quickly superseded by a simpler solution. The high-frequency arc-lamp power supply works well and does not have as much complexity as envisaged - this is often the way thing turns out that is the final simplicity conceals the effort necessary to achieve that.

Heath Robinson - drew cartoons of elaborate machines but engineers make a machine to function well in 
an uncomplicated way from a range of imprecise materials and parts. https://www.digitalartsonline.co.uk/

It is not an uncommon problem for small companies and individuals to have difficulty in getting support from big companies on products those companies make.  Many parts made by big companies may not be available to other customers but it can be worth asking about them.  But also many very big companies are particularly supportive of small companies and individuals.  There has been a positive change over the past 20 years with big company's helping small company's and individuals much more although some never stopped and have always worked that way.

Spectral-lamp
Laboratory Spectral-lamp-unit

Half-bridge - switch mode vapour arc-lamp power supply;

Based on a successful design but with changes and using newer parts than are fitted in the actual lamp power supply in the early 1990s.  The lamp and power supply is enclosed within a metal box with metallic contact on all edges to the lid.  There is an aperture for the light to exit - this provided the lowest emissions in conducted EMC test up to 30MHz at that time.

Since working on this project I have worked for automotive contractors one of those had a test chamber, LISN (line impedance stabilisation network) and a large field test site.  It is possible to use a LISN and analyser up to 150MHz - although un-calibrated this will making initial testing more practical.  Using an oscilloscope is essential first step look at the diode overshoot which is the diode turn-on time and may cause EMI.  Measure the width of overshoot if it is say 10nS (half a cycle) then any related to the diode emission is likely about 50MHz.

The most significant thing about using a LISN is that you do not just measure conducted emissions but you also get a good indication of potential radiated emissions without needing to leave your workspace and equipment.

The main power supply AL-00026-01a.  The power supply has components rated for 200W - 400W 

 output although the lamp power is much less.  C4 was fitted on the PCB but on reflection may have 

been counterproductive to fit? This is not complete or final but this circuit works very reliably.

C16 & L2 form a tuned circuit to creates the high voltage necessary to strike the vapour arc-lamp's arc.  The frequency first reduces to below resonant frequency (about 50KHz but 70KHz and 100KHz also works, that is C6 changed from; 22nF to; 10nF then 4n7 respectively) to maintain about 150mA until the frequency drops to reaches 20KHz.  The circuit then runs at 20KHz and the current in the lamp is controlled at 1A by pulse width modulation control.

The current mode switch mode power supply controller works well in this simple current output design providing some of the current regulation.  There is a conventional slower outer control to set the lamp current accurately fast as practical to minimise mains frequency modulation of the light output.  The operation is un-symmetrical the negative pulse width is shorter, higher and set by the controller, the longer positive pulse completes the remainder of the period of oscillation.  The controller also includes over voltage shut down in case the lamp is open circuit such as if it is not fitted.  The overvoltage is sampled on the positive peak and the limit is set at about 1,500V peak, the peak voltage will be a little higher than this, so the capacitors are rated at 2,000V but the inductor's voltage rating would need to be confirmed.  The lamp's light output and life appear to be very good is more uniform light noticeably at each electrode where you see fluctuation and ran reliably daily for at least a decade.

The circuits operating frequency ramps through L2 & C16's resonance which causes the high voltage necessary for the lamp to strike.  Otherwise, the circuit operates in a similar way to how the conventional 50Hz choke ballast above works.  When the lamp is at full temperature C16's purpose is mostly complete although it may provide some filtering of RF emissions.  L2 provides a 1A average current source for the lamp.  Spectral Lamps require 10 to 60V at 1A RMS depending on the lamp metal.  The lamp was plugged into a B9A valve base, called PICO 9.

Low Voltage Power Supply

The circuit below provides low voltage power for the controllers.  This power will shut down if the lamp does not start after a short time.  The mains voltage PTC probably should be rated at about 2-5mA in order to be overload and shut off but allow a reasonable running time for the lamp to start.  This function is not fully optimised but is a design proposal.

The running current needs to be recalculated so that PTC be a better choice in view of the controller may be lower powered, the gate driver higher powered and the gate power probably lower power despite the higher operating frequency?

   AL-0026-01C low voltage power supply with timed shut-down.  The transistor type 

and other parts are not specified the part values shown are place holders only.

The next revision will have Power Factor Correction (PFC) that could also provide low voltage power for the arc-lamp section.

The existing low voltage power supply above in which the lamp current sense transformer provides continuing power.  The start-up PTC, R13, is under-sized so that the initial low voltage section's power is only provided briefly until the PTC shuts the power down.  All the power is turned off if the lamp fails to start within a short period when the PTC gets hot and goes high impedance.  A disadvantage with this function block is that some power has to be wasted in the zener diode.

From discussion below an additional thermistor, R14 has been added under the top transistor to shut down and protect the power supply.  This part could have an alternative jumper by-pass in case a heat-sink needs to be fitted instead so that there is a flat area under the transistors for a heat-sink.

Assessment of the original arc-lamp power supply;
The power consumption of the whole power supply was fairly high but this did not matter, the lamp is an excellent source of flicker-free and low ripple light.  The main inductor which is about 40 x 40 x 40mm depending on which was used ran hot or very hot depending on which one was used.  The transistors were each mounted to PCB mounting heat-sinks rated at about 15'C/W each and they ran cool.

• Increase the size of the inductor or increase the frequency of operation to reduce power loss or at least to reduce the inductor's temperature rise.

• But regulations were coming in that required electrical equipment to have power factor correction if the power was greater than 60W.  In addition, sodium street lamps if only sodium light was required is better priced and lamp life with this option is longer but street lamps use more power and produce more light than a sodium Na Spectrol lamp.

Earlier design often destroyed transistors at startup;
When the arc is struck the output transistors were often destroyed.  My design assumption was that this would not happen because there would be enough stray inductance for C16 discharge current to be limited the current adequately was wrong.  Secondly, the circuit for the arc-lamp current path was not near the power transistors and I could not see a path on the PCB that was nearby either to explain this.  Although the current sense transformer did carry the high discharge current nothing was damaged in that circuits path.  L1 was added although in that location about 4u7H or 220mR wire wound resistor adequately resolved the issue -  for such a small impedance to be significant then evidently the current pulse was surprisingly high.

The failure resolved in conclusion;

I could not find a reason for the transistors failures but the fix is a good practice anyway because that fix minimises pulse current and therefore EMI and is necessary in order to reduce the tuning capacitor C16's peak current and therefore reduce components stresses.  It is reasonable to limit currents when possible even if it may not be apparent why it is necessary.  The solution was to grab at straws and try limiting the arc striking current.  That is I have not found the path of failure but in future, I can take mitigating measures.

Now that I have a better feel for these things that I had not been aware of, next time it will be easier to design those mitigating measures first.  Then consider leaving them out later or testing design margin later.

Op-amp unstable;
There is a dual op-amp (LMC6032IN) in the controller section used to produces the frequency ramping in order to strike the lamp that was unstable.  The instability may have been due to the electromagnetic emissions from the power inductor.  Alternatively, this amplifier was likely to become unstable anyway if the output load was greater than 80pF and the op-amp had a light resistive load.  A workaround was to add a 100uF capacitor to the output - this method of overcompensating is not documented in the data-sheet for this op-amp.  The data-sheet recommends that the output be loaded with a pull-up resistor to draw more current.

----- Select another Op-amp and move the op-amp to the bottom layer away from the power inductor ----

Changing to an SMT design may also allow us to separate the input inductive filter from the output choke by distance then put this op-amp and other electronics on the bottom layer thereby screening that electronics with the PCB's 0V plane between.

Power Input filter;
The input filter circuit with the safe X and Y class capacitors have changed a little since about 1985 other than to become required standard practice to fit rather than being optional. Some of the parts shown will also need to be reviewed.

 Input filtering AL-0026-01A - there is only a single point connection 0V to Earth connection 
although bonding at many points would be better - this cannot be done for safety reasons 
one single point RF connection is achieved by commonly available safety approved type 
Y-class capacitor.  In any case, star point strategy works fine with care.

The input filter circuit is compromised for safety in that there is a single bond 0V to Earth via the Y class capacitor.  Further down this page another design proposal is more efficient and includes an output transformer.  With that design one of the lamp supply voltage connections is connected to earth for potentially lower EMI.

The Earth and Neutral connections may have two resistors in series of a safe type and a safe current connected between them in order to discharge C3 thereby minimising the tiny electric discharge that a persons touching the plug might get.  The disadvantage is that safety is slightly compromised in the unlikely event that both resistors were to fail short circuit.  The discharge energy, in any case, is safely within the standards without such a resistor.

------------------------------------------------------------------------- 

  

Developing and refining this 1990's design

The circuit near the top of the page is very similar to the working original circuit is the basis of an improved design.  There are a number of incorrect parts shown they are place holders in order to put down and develop the ideas without being distracted by detail.  Polypropylene and Polycarbonate power capacitors will need to be changed to Greener RoHS compliant types.

The earlier working design included a full-bridge 24V DC, mains voltage full bridge and also some other mains voltage circuits.  The design was not carried out to this degree of detail but assumptions were made - I needed to see the circuit work before I could fully appreciate the characteristics of the metal-vapour-arc-lamp.  The lamp data-sheets have very little detail but in any case, had no detail about operating the lamp at high frequency but I had anticipated that the lamp starting would be easier from prior play (not work) with fluorescent tube.  Therefore more design parameters were discovered by playing with partly working ideas.

 General block diagram;

AL-0028-06C Block diagram has an additional block - the Power Factor 
Correction.  An aperture for light to exit which introduces an EMI issue is shown.

The block diagram shows some changes from the initial design and is similar to AL-0026-01C but the power factor circuit will need to be developed.

Output transistors - first look;

The original circuit used IRF841 transistors plus a PCB mounted heat-sinks each for the half bridge output.  Although the newer transistors below also include an avalanche diode and the diode has a reasonable switching time but the switching losses could be a significant factor.  That original circuit had a transistor current limit of 4A.

• A rough estimate for candidate new transistors power can be done using Ron.

• Additionally comparing transistor Qc (gate charge) will give an indication whether the operating frequency can be increased without switching losses becoming excessive.  The parameter Eoss = 1.1uJ at 400V (IPD60R600P7) may be a better figure but they are probably linked in any case it also depends on the topology chosen.

• IPD50R280CE  Ron= 280mR, Qc = 32.6nC -- this part is shown in the circuit above but was not modelled.

• IPD50R500CE4 Ron=500mR, Qc = 18.7nC

• IPD50R950CE  Ron=950mR, Qc = 10.5nC

• There is a trade-off between Ron power loss and switching losses.  Qc is approximately correlates with switching losses.

• The design below as refined largely mitigates losses due to capacitance although the parameter Eoss is not mitigated and is the minimum number of joules per output transition to estimate power loss with.

• The parts above are not recommended for new designs and in any case, they are not optimal because they have higher capacitance including gate charge and therefore higher switching losses compared to newer parts. 

• The substrate diode in the new parts below turn-off a little faster.  This parameter is not too important because it is unlikely to limit the minimum pulse-width in considering operating at a higher operating frequency.  See simulation below.

• These parts may be cheap and easily obtainable or already used.  Best to only use them if they are already used elsewhere.  Using what is available rather than is desired is often the only option in a new design.

• If the transistors are soldered to PCB with no additional heat-sinking then the package dissipation is 62'C/W.  That figure depends a lot on the mounting orientation of the PCB.
  

The top transistor's Drain is soldered to an area of copper for additional heat sinking.  I have included a lot of decoupling to minimise EMI radiated by this pad which is connected to the positive supply.  Discussed below a PTC thermistor to sense over temperature and shut down the controller could place in this area.

The bottom transistor's Drain is soldered to an area of copper for additional heat sinking.  But the copper area will cause EMI radiated and it will be necessary to experiment with snubbing and switching time to minimise EMI adequately.  The bottom transistor conducts high and shorter current pulses (although the maximum current is not as high as might seem at first).

https://electronics.stackexchange.com/questions/26783/how-do-i-determine-the-area-of-copper-needed-on-a-pcb-to-provide-adequate-heatsi?utm_medium=organic&utm_source=google_rich_qa&utm_campaign=google_rich_qa 

This example with 50mm x 50mm one side of the board but with a smaller transistor would improve the thermal cost-efficient to better than 35'C/W.

----- Return to the transistor power selection later -----

Simulation to show basic circuit operation;
Simulation adds a step but gives me more insight into the operation of the circuit before proceeding to have a PCB made. This model will confirm or remind me of assumptions and calculations made previously.

Simetrix 8.1 simulation of the output choke/filter and the lamp represented by the bridge rectifier 
and voltage source as a clamp.  AL-0026-02b.  Bottom right is a model for the arc-lamp.

• The transformer is; 1:20, therefore R2 = 20R give 1V per amp average.
• AL-0026-02b lamp; 15V
• The simulated mean voltage is; 1.04V which corresponds to; 1.04A
• The simulated current in the wire is; 1.31A_rms 
• The current is too high but if R is changed to 27R which is a closer approximation for a triangle or sawtooth waveform the current will be about 8% above the desired current.

• AL-0026-03C lamp; 100V
• The simulated mean voltage is; 1.07V, which corresponds to; 1.07A
• The simulated current in the wire is; 1.53V_rms 
• The current is too high but if R is changed to 30R which is a closer approximation for a triangle or sawtooth waveform the current will be about 8% below the desired current.

• R3 damping resistor has been increased to 220R and C2 reduced to 10n.

• The lamp type simulated is a Sodium lamp but this model using a voltage clamp for running mode is not good.  Other simulations below of the lamp running normally use a resistor instead.

After modelling R3 was added to reduce L2 & C2 ringing, there is a small oscillating at high frequency observed in the model at the lamp supply.  The power loss in R3 will need to be calculated but the simulator can do this.  C1 has a reactive voltage of 130V pk-pk and does not have a too severe effect on the output current that could have prevented the current mode SMP controller operating if the slope were to become negative consequently.  But if C1 were reduced to 220nF the reactance is significant and at 100nF the pk-pk is ~3.5KV and prevents a positive current ramp required by the current mode controller occurring.

R3 to lower the Q of L2 may not be necessary because the lamp characteristic when running appeared to be resistive.  In any case a soft magnetic should be chosen which has a low Q and is for such filtering likely required when the lamp starts and is warming up because the noble gasses are conducting and the lamp is likely to behave like a spark-gap transmitter.

Waveforms for the above model AL-0026-02a, 15V out.

• Blue - average voltage (current) about 1.2A in the lamp.  Using average current rather than the RMS current worked well but the current is lower than specified for the lamp.

• Mauve - Driver output current is about +1.5A to -2A and is a clean waveform.  The bottom transistor is on for 5uS then turned off and the top transistor is on for 45uS.  This waveform is conveniently inverted so therefore the waveform's don't clash., The inversion may be due to the direction the wire is drawn in the diagram? 

• Cyan  - Lamp current this was oscillating a lot until R3 was added.  R3 lowers the Q of L2.  You can see some high-frequency oscillation which I think is due to the limitations of this model.

At 100V output, the peak drive current is -1.3A +3.7A and the high-frequency ripple 
is higher.  This was with the bottom drive transistor on for 8uS.  AL-0026-03a

By inspecting the waveform's it looks like the current circulating in the top transistor is about 1 to 1.3A when the current is correct for the lamp.  The circulation high side current in the lamp is beneficially fairly flat. Increasing C1 to 10uF reduces the bottom side transistor peak current to 3A - this is useful but there may be other trade-offs?

It is evident that as expected the power choke L1 is undersized therefore the current is higher and the on-time is shorter than ideal.  This choke was chosen based on availability and to make the design more flexible.  Later the design review will also look at using a higher voltage lamp and also adding a step-down transformer.

Modelling a switch-mode power supply without using simulation tools is also relevant;
In a conventional switch mode power supply the current rises in the inductor are;  di = V.dt/L so the current increases with on-time so the power increases with the on-time squared.  Consequently to increase the power transfer (the voltage Vs is constant) the control loop will increase the on-time by t².

In this case, though the longer top-transistor on-time is shortened and the power is reduced so with close to 1:1 PWM the change is roughly linear. The approximate square law assumption is about correct for lower power lamps running with a high mark-space ratio.  So this lamp control is more complicated than a conventional SMP.  There is a benefit for a low power lamp the control loop gain does increase so much compared with a linear power supply and the control loop can be faster but still be stable than might be guessed at first.

Consequently using an estimation of minimum on-time based on square law in order to consider increasing the operating frequency is a useful modelling exercise.  The next output transition (minimum PWM on time) is determined by the time between the current flowing into the substrate diode in that transistor (current flowing out of the inductor but not conducted by the MOSFET) must be at least tRR (~200nS).

Conclusions for the simulation of general operation above;

Lower the Q of L2 by adding R3 to reduce ringing; Add R3 but do not fit this resistor.

It is unlikely that this extra design step would be assisted in the design process application notes generally do a good job of explaining standard topologies.  But modelling the power in the transistors is likely to be beneficial - I had been advised that would be the case!

Output transistors 2;

• IPx65R310CFDA - this part is recommended for new design.

• Ron = 310mR (725mR @ 150'C), 62'C/W (35'C/W 6cm² 
   copper), Qc = 41nC (6.4Vplateau @ ID= 6.6A, 480V), diode tRR=95nS @ 6.6A (this is better)

• Various figures in the data-sheet suggest that the dead-time (when both transistor are turning off and turning on) could be reduced from 500nS to 100nS.  A the model used results in 400nS. 

• From modelling below the transistor's power consumption is much higher than anticipated.  Consequently using a copper area on the PCB would not be adequate with this transistor.

• IPD60R600P7 - the power in the transistor comes down drastically to under 1W and 1.6W at 15V out.  That power is shared between the switching and Ron losses equally. 

• tRR=160nS at 1A - Slower than a similar voltage fast discrete diode and much slower than a lower voltage Schottky diode.  This parameter is not significant.

• ID max is; 16A. 

• MOSFET ruggedness; 80V/nS, Diode; 50V/nS. 

• Diode commutation; 900A/uS - This figure seems to be used to calculate Vsd(peak) the substrate diodes turn-on time and voltage and snubber capacitor will limit this.  I have found it difficult to confirm that that is what this parameter is used for but there is no other use for the parameter.  Negative reasoning is not sound but is the only option sometimes.

• The example in the data-sheet does not show the difference between the board the copper area being below which is poor, above which is good or the board mounted vertically which gives 50% more cooling.  Also exposed tin, which oxidises mat white is better than green resist on top.

• The DPAK package heat-sinking is; 62'C/W but with a 6cm² area of PCB will reduce to 35 - 45'C/W (maximum from the data-sheet) The thermal time constant can be estimated from diagram 4 in the data-sheet the curves level out at 10mS which is about 7 time-constants.  1.5mS.

In reply to a question about thermal time constant to Infineon, the advice was to use the spice model even though it is slow but taking the figure from diagram 4 is correct.  I'd say do that despite the contrary advice it is much quicker and clearer to use the figure from the diagram.  I was also trying to find the time constant of the PCB copper as well I may return to that - if I was working with a team I may ask a mechanical engineer or physicist but in any case, the copper mass is very little so the figure will be very small.

Simulation with power transistors;

The snubber

• 220p + 27R and other similar values the optimum dead-time is at 200nS.  Found by experimentation with the model.  The peak current due to the snubber (which is all the current external to the transistor) is not normally reached but is because the current flows into the inductor but is; 16A = 400V/25R but ID max is 16A but I've added a very small margin. 
• It turns out that the snubber may be better used to limit dv/dt so that losses due to the miller effect in the transistor do not occur when running normally.  Therefore R can be reduced to 0R.

• It, therefore, turns out that the transistor turn-on time is a better limited by high-value gate series resistor than by fitting a higher value snubber resistor.  A small value snubber resistor value of 0R47 is likely to be required to minimise ringing.  This will be considered below (AL-0028-??).

The simulation below now consider the switching power but does not include change with temperature increase.  It will, therefore, underestimate power consumption because Ron increases with temperature rise so, therefore, the power dissipated increases with temperature. 

AL-0026-04b 20KHz and 71KHz starting, The simulation using the newer gate driver (IR2143).  
C1 and R1 in practice would be transposed in order to minimise the physical area of components fluctuating at high frequency.  The snubber R2 and C3 are placed optimally for the same reason. 
R4, R3 and R6 are necessary to protect the driver as Infineon; DT-97 3 art "Managing 
Transients in Control IC Driven Power Stages", Method B.

Typical output power losses simulated in the power transistors coming down from 4.5-8W to 0.5-1.5W per transistor after also trying other newer transistors.

• The gate driver model used is simpler than using the spice part.  In any case, the spice part is too complex for the free version of Simetrix used.  In any case, this model is not compromised by being simpler but probably is faster.

• IR2143 gate driver - Increasing the drive current to 1.4A/1.8A simulating and the power out 15W (<1W) and 100W (<1.5W).  Note I needed to run the model for a longer time to get a lower more accurate power figure.

• The model for the metal vapour arc lamp is changed to simply a resistor.  This looks like a better model for the lamp fully warmed up.  Two resistor values 15R and 100R were modelled with different pulse width settings in order to produce the desired 1A.  (this is not the final solution the snubber, gate drive and the reasoning will change)

• The snubber resistor is also reduced in value as a consequence of observing that the maximum current is limited to 4A and that would occur at 50% of the supply voltage.  

One of two waveforms for 15R = R1 output the other waveform for R1 = 100R is not shown but is similar.

The anticipate high power peaks at switching time can be seen on this graph.  AL 0026-04b. 
The power R2=47R not as shown and the power in the transistors is a little higher 
consequently but maximum ID is not likely to occur.  Dissipation is now spread equally 
between the switching and Ron losses.  The biggest loss is in the snubber which is likely to 
be required to minimise EM emissions.  The power in each transistor is now about 1.5W.

- Lamp voltage (R1)
- Q2 power  
- Q1 power

Modelling the power output transistors;

Advice from PCB makers and assemblers on the exposed metal area for heat-sinking;

• Gardner Osborn advises that it is possible to have a copper area exposed so that there is no green solder resist mask but it is not possible to have this area black oxidised.  The inner layers are black oxidised but so it would be necessary to cut an aperture in the top and bottom layers so that the inner layer is exposed.

• Wilson Process Systems say if it is possible to have a black oxide exposed copper surface it is likely that that would only etch back a few micrometres of the 70um copper thickness but they gave me the PCB's supplier name to check with.

• I received a similar answer from Minnitron Ltd. who is the oldest PCB manufacturer in the UK.  Black Oxide used to be used but it probably would not fare well exposed.  Generally, transistors are mounted to a Lead-free hasl or Ni/Au finish on exposed copper lands.

In the 1990's I found that a PCB with an area of copper for heat-sinking was improved by having the green resist left off of the heat dissipation area of a PCB.  The area of PCB with the exposed tinned copper quickly changed from silver to tarnished mat grey improving the power dissipation.  A variation like is still possible.  Particularly leaving the black oxidised copper exposed - doing this could increase the cooling by 50% provided a thick enough area of copper can be used such as 70um thickness rather than standard 35um thickness.

   Thermal model AL-0026-05b - this model is slow and does not add any useful information that you can't resolve 
more quickly with a calculator.  Gate driver protecting Infineon; DT-97 3, Method A is shown D1, R7 and R8.

Thermal modelling needs to be done using Ron increased to a maximum at the highest operating temperature but to start with 150'C (maximum) can be used.  From the data-sheet Ron will double but if we use 25'C for now and keep in mind that the temperature rise due to power dissipation may increase by 50% to 100%.

• See later that further circuit changes result in a reduction in power dissipated proving adequate margin.  It is therefore not necessary to work with this thermal model any more.

The gate driver protection; - was not used before and the design was robust;

• Without the protection circuit, the lower transistor or substrate diode must have turned on before the top transistor swung below 0V by -20V or so.  In any case, the snubber may have slowed the dv/dt enough to prevent such a severe negative voltage occurring?

• Method A - the top transistor is likely to turn on and draw excessive power when the inductor current drives it negative (below 0V) during the substrate diode turn-on period, this period is very short. (AL-0026-05?)

• Method B - this is fairly good but a series resistor needs to be added to any input to the driver (that would be in effect in parallel with AL-0026-04b, R6) 

• Consider not fitting the recommended extra circuit - the driver's internal circuitry protects to -25V. The snubber will, in any case, give the MOSFET a little time to switch, about 7nS or 7+6nS with the extra diodes D3 and D4 if they function as guessed (AL0028-01 below) but checking this there is not adequate protection particular when the choke is running full power at resonance (5KW circulating briefly perhaps). 

• The addition of D4 should work, I understand that the Infineon C7 MOSFET gate threshold is symmetrical either polarity, so the MOSFET will turn on after a delay when Vsd(peak) exceeds 5V or so the gate plateau level.

• In normal operation, negative undershoot is not or is barely significant in the simulations and the waveforms look clean.

• Even with the methods recommended there is still a reliance on diodes turn-on quickly. Those diodes are smaller or lower voltage so they will, in any case, switch-on more quickly than the transistor substrate diode. 

• There is another gate driver IC's that the high side driver can withstand a high negative voltage swing such as made by Texas Instruments UCC20225 and also Infineon 2ED***** series. See below.
  

• Vds(peak) the substrate diode turn-on voltage is shown in a figure but not given. 

• For C_snubber = 100pF at 1A 
• A commutation value 900A/uS may be the correct value to calculate Vds(peak)?

• At 1A is; 1.1nS = 900A / uS

• Capacitor value and chose current will be increase in tandem.

• i=C.dv/dt therefore; dv = i.dt / C

• 11V = 1 * 1.1E-9 /100E-12 ---- This working out looks okay?

• Vsd(peak) needed to be worked out anyway because the maximum voltage stresses that the transistor work to needs to be known. 11V is fairly trivial it turns out.

Snubber to reduce likely VHF EMI and minimise the transistors switching losses;

Summary of what follows; The method recommended by Infineon for C7, MOSFETs is to add a gate series resistor and use the MOSFETs output capacitance in a controlled way. This is for a different power supply topology and has not been done in that way in this case.

• The application note for the C7 MOSFET also recommends fitting a series beed inductor in the gate connection. This has not been simulated. No ringing has been observed in the gate simulation and the high gate turn-on resistor has most likely resolved this issue?

The first circuit shows a snubber circuit formed by R3, C7, the same snubber is simulated in AL-0026-05?, R2, C3. The snubber will consume some power but the circuit is intended to slow the current rise in the transistors substrate diode in order to minimise the generation of high frequency (that could be 50MHz) EMI. The snubber may also be necessary to protect the transistors and the driver? The EMI is due to substrate diode's turn on time. There are two other figures given for the diode;

• The design reasoning turns out to be less than optimal. R3 wastes power and does not limit dv/dt until a voltage is developed equal to the inductor commutation current. Consequently, there are some losses due to transistor miller capacitance.
  

• Diode dv/dt maximum is 50V/nS. (diode is the limiting factor); At start up the inductor current is 5A (3.5x running frequency for starting)

• i = c.dv/dt, Therefore c = i.dt/dv
• 100pF = 5 x 10e-9 / 50 - the value would be halved due to the transformer (AL-0028-01x). below but the margin is small.

• Calculating the snubber capacitor to slow the output transition in order to avoid miller capacitance in the transistor being turned off from causing that transistor from being turned on and causing extra power wastage. During normal running alternately a transistor is turned off when the inductor current is maximum so consequently the voltage would swing rapidly to the opposite polarity without a snubber capacitor to slow the dv/dt rate. Called flyback or in this case is called commutation current. 

• IPD60R600P7 has a minimum gate threshold of 3V (worst) and series resistance of 6R3 (typ)

• The gate driver with the extra diode shown in AL-0028-01x below ~1V

• Maximum gate current that will cause power losses due to miller capacitance is;
• Gate current that will not turn on the MOSFET is; <270mA = 3V-1V / (6.3R + 1R)
• Maximum dv/dt is that can be handled without gate current exceeding 270mA is;
• Using Qc=9nC and Q = i * t, rearranged for t. then t = Q / i.
• 33nS  = 9nC  / 270mA - this figure looks too low.
• Output snubber capacitor;
• i = C.dv/dt As above therefore; C = i.dt/dv
• 82.5pF = 1A x 33nS / 400V
• 83pF per 1A.

• It turns out that the problem is solved for us the data-sheet gives figures for one condition; 400Vs, ID = 1.7A and Rg=10R then these figures are given;

• Rg proposed is <10R so the switching times should faster which is better but there is an additional diode drop which reduces the drive and offsets some befits.

• Delay Td(off) 37nS -- this figure is not required.
• Fall time Tf 19nS

• 40nS = Tf x2 (estimate for 100% rather than 19nS between 10% & 90% points)

• i = C.dv/dt As above therefore; C = i.dt/dv

• C required to limit the output fall time; 200pF = 2A x 40nS / 400V
• Or 100pF per 1A. 

• We can see that the 40nS from above gives a dv/dt;
• 10V/nS = 400V/40nS this is well within the substrate diode rating. 

• The model is changed below to limit the transistor's turn on time by using a high-value gate pull up resistor.

The possibility is that by increasing the snubber capacitor further that the circuit's output transitions be slower. This potentially gives similar benefits as using resonate mode in making the switching losses and EMI low. But not to increase the capacitance so much that there is very little control range as I have considered for something else previously.

• From further down the page it can be seen that a disadvantage could be that during the starting and warming up periods a larger capacitor will need to be driven by more transistor current. See waveform AL-0028-02? and also below.

The snubber conclusion;
There is more modelling below but the outcome has been placed in this section. Resulting in improved theoretical design efficiency. The capacitance values need to be scaled for the maximum inductor currents when normal running mode, the last equation above turns out to be the significant one. 100pF per 1A maximum current. It also turns out that it better to allow the snubber to be under-sized for the brief starting mode in which the current is highest so that the warming up mode be more efficient.


Power transistors;

This project has been entered in a contest run by Infineon although I do not intend making a prototype. I have used technical support of various manufactures in order to improve this design exercise.

The improved efficiency switching by added output with selectable capacitors see below instead of  a snubber will the feature of a project posted at; https://www.hackster.io/

Many of this batch of large metal TO-3, OC28 and OC35 transistors pictured was made in 1969. 
These would be the state-of-the-art in the 1950s. Small surface-mount transistors;  IPT60P028G7  
Infineon, pictured in the foreground, is currently claimed best in their class power MOSFETs.  
Placed on their customs delivery note. This project specifies lower power MOSFETs from this series.

Bipolar-Junction-Transistors cannot be operated at maximum current and voltage even briefly but this limitation does not exist in MOSFET devices. MOSFETs can be operated at maximum current and voltage briefly. Newer Germanium power transistors were very good but they had limited applications when they were withdrawn in about 1977 by Motorola for example.

IPT60P028G7 Infineon's best in the class power MOSFET, 600V 75A. 

Two higher power germanium 1960's technology transistors could be used in a 50Hz, 300W, 24V to 240Vac inverter very efficiently with virtually no heat-sink. Silicon BJT's had a higher saturation voltage so were less efficient consequently they were not always a replacement. Adequately power rated MOSFETs that would have the comparable low loses did not exist for another 10 years and were expensive. At the same time 1976, SG3524 SGS the first switch mode power ways supply IC became available and these would never be used with germanium power they required higher frequency silicon power, to take advantage of smaller magnetic parts.

Review circuit topology and alternative strategy's

• Bought in power supply with 48VDC output design for lower voltage lamps up to 15V only but keep the same topology - this is easy and suits a low volume manufacture design because much of the compliance for CE marking is purchased with buying the power supply. The size of the power inductor reduced. The transistors can be 100V optimised for high-frequency operation thereby reducing the power inductors size further.

• A variant of this used a 24V full-bridge SGS MOSFET stepper motor driver successfully. The new drivers have thermal protection built in that would be an advantage.

• Add a step-down transformer so that the current in the transistors is reduced, the power in the inductor and all other power circulating be further reduced by operating with a smaller input-output difference.

• Spectrol lamps are expensive but the following sodium lamps are often used instead they are also higher voltage, lower current conveniently; NAV-T 50 W SUPER 4Y or NAV-T 50 W SUPER 6Y OSRAM, 50W, 86V.

• Increase the supply voltage to 500V to suit the transistor rating 600V suitable electrolytic capacitors are available although 450V capacitors are cheaper and more common. Increasing the supply voltage also makes the PFC control easier because the operation range is smaller due to the greater headroom above 385V rectified mains voltage to 450V instead of 385V to 400V that used to be the case in the 1990s. 

• 390VDC out from the PFC is still common and it gives the transistors better over voltage margin allowing a smaller snubber capacitor and there fewer losses in that component.

• Increasing the supply voltage does not reduce efficiency by much but could allow the use of an off-the-shelf transformer to be used. In any case, a custom transformer is an easy option so any benefit even for prototyping is likely to be marginal.

• A single transistor steps up forward converter that would require the design of a leakage transformer (loosely coupled) or use a transformer and choke. This option was tried but it is likely to be less optimal than the circuit chosen. What has been developed subsequently resonant to start up works well so the forward step up solution need not be developed further at this stage?

• The circuit was tried but not refined using a  GTO and LT1070 but these parts are now mostly supperceeded. The circuit was robust but did not function fully just proving that this type of thyristor is also very robust.

• Operate at or near resonance frequency so that the startup frequency need not be varied.

• Risk some units will not start lamps due to component tolerances results in the frequency being too far away from the resonant frequency to reach the arc-lamp striking voltage. 

• The risk can be mitigated by adding but not fitting components to the PCB but adding them later with little re-working of finished PCBs.

• Don't bother the extra circuit is fairly modest.

• The losses are higher than optimal at starting because of the higher circulating current. This may be offset by not having a ramping frequency which may or may not cause high EMI than with a fixed frequency. 

High voltage reservoir capacitor;

Now that the output transformer has been chosen a large electrolytic capacitor needs to be selected along with a supply voltage at <100W running and warming up. Looking at price 400V to 450V at >33uF and > 250mA ripple. 47uF 450V is reasonably priced. The assumption is that the starting current of <5A is only brief and not significant.

Location; common to the output of the PFC and the HF light source Half-bridge output.
F=100Hz, C=33uF (400V or 450V)
Impedance = 1 / 2.pi.f.c
50 ohms = 1 / (2 x 3.14 x 100 x 33E-6)

The ripple voltage when running will be ~25V. (100W lamp 400V).
This capacitor can handle the maximum ripple current required at the lowest price. May be formed by two capacitors for lower EMI.

Revised Snubber function;
Transformer added to the output improves efficiency and the revised gate drive reduces transistor losses further.

The gate drive in the circuit below is a better solution than those above. Significantly the snubber resistor is zero or much lower in value and the transistor ID limiting is achieved by slowing the transistor turn-on time. The potential benefits become apparent when the start-up conditions are modelled below.

Simulating with transformer output added;

• Gate turn-on resistor; to give transistor turn-on >40nS;
• Qc = 9nC, no minimum or maximum is given.
• Q = i.t, therefore gate current less than; i = Q/t
• 225mA = 9nC / 40nS

• Vs(maximum) -Vth (but gate plateau voltage may be better)
• 12V = 15V - 3V
• Rg = Vs - Vth
• 53.3R = 12V / 225mA

  AL-0028-01E transformer coupled output 20KHz running and 50KHz starting. 

This circuit does not have gate driver protection but potentially the single resistor
connection to the gate should make a good PCB track layout easier to achieve.

Modelling with transistor shown IPD60R600P7-L0 also significantly improves efficiency. Also, see below. R2 is now used to minimise ringing and will be reduced in value further (0R-0R47) but ultimately is determined by EMC performance when tested. The transistor power comes down below; 600mW

Conclusion for the revised snubber function - to improve efficiency.
The conflicting outcome so a compromise must be found;

• C3 seems to need to be higher in order to optimise the power supply at starting. But it is not necessary to minimise losses in the power transistor during starting much because the unit only operates in this mode briefly. Later it will be found this is not practical.

• Later it is found that C3 needs to be reduced during the low current warming up mode where the starting up capacitor's resonance is still significant. The current direction varies and C3 is a disadvantageously increases losses due to transistor switching on when the voltage is not zero.

• It is beneficial to make C3's value higher in normal and starting modes up to 1nF. But instead, even lower values don't work during the warming up period which is a number of minutes. Instead, it looks necessary to try C3 reduced below 100pF if higher operating frequency is to be used.

With the conflicting simulation that in conclusion C3=330pF should be reduced further or else the running frequency be reduced. This simulation works well for normal running mode only and may not be over-all the best compromise.

Waveform simulated from the preview circuit above;

 AL-0028-01E 20KHz and a 50KHz preview of the outcome of the simulation below.
 Transformer coupled output. A one-half cycle is shown 500nS delay, 20KHz.

Transistor current waveform shows inductor current flowing into the substrate diode. The transistor is turned on and that current will then be shared with the transistor. During this period until the transistor is turned off is a minimum period of diode tRR it can be seen that there is plenty of time for the diode. The IPD60R600P7 transistor dissipate < 600mW.

• Green - Half-bridge output voltage The snubber capacitor slows the dv/dt. No undershoot can be seen.

• Green - Bottom transistor current some ringing can be observed.

• Blue - Gate Drive turn-on delay can be seen at the simulated current that the delay could be shorted.

The transistor losses are similar at; 20, 50 and 100KHz but increase a lot at 200KHz in normal running mode. Therefore 100KHz is probably optimal normal running frequency.

----- It has turned out to be difficult to repeat the results above partly because a long enough run time is difficult in Symetrix ----

At 100KHz running normally C3 can be reduced further the worst case highest current is for a 10V lamp and the worst case lowest current is 1:1 output for the maximum lamp voltage. The simulation tends to fail to complete the simulation will complete if the duration is reduced but this gives a higher power figure. Simulation time of 200mS is about right but fails to complete so a delay of 20mS and completion at 25mS works and is used instead. R2 = 0.47R damps current oscillation well in these simulations.

D3 & D4 were added as diagrams below (brackets are used for a different simulation time);

• At 100KHz, dead-time 200nS, 47pF, 47Rg the transition time is 45nS 45nS; the transistor power is 2W & 2W

• At 100KHz. dead-time 200nS, 100pF, 47Rg the transition time is 90nS 65nS; the transistor power is 1.4W & 1.4W

• At 100KHz, dead-time 200nS, 100pF, 100Rg the transition time is 80nS 60nS; the transistor power is 1.5W & 1.5W (10mS - 15mS)
  

• At 100KHz, dead-time 200nS, D3&4 not connected - 100pF, 100Rg the transition time is 80nS 60nS; the transistor power is 1.5W & 1.5W (10mS - 15mS)
  

• At 100KHz, dead-time 200nS, 100pF, 220Rg the transition time is 80nS 60nS; the transistor power is 1.5W & 1.5W

• At 100KHz, dead-time 200nS, 150pF, 100Rg the transition time is 120nS 85nS; the transistor power is 1.3W & 1.3W 

• At 100KHz, dead-time 500nS, 150pF, 100Rg the transition time is 100nS 80nS; the transistor power is 1.3W & 1.3W (10mS - 50mS)
  

• At 100KHz, dead-time 300nS, 220pF, 100Rg the transition time is 120nS 150nS; the transistor power is 1.1W & 1.1W

• At 100KHz, dead-time 300nS, 220pF, 220Rg the transition time is 160nS 160nS; the transistor power is 1.1W & 1.1W

• At 100KHz, dead-time 500nS, 330pF, 100Rg the transition time is 180nS 220nS; the transistor power is 950mW & 950mW

• At 100KHz, dead-time 500nS, 330pF, 220Rg the transition time is 180nS 220nS; the transistor power is 1W & 950mW

The transistor's effective output capacitance is figure in the data-sheet is 150pF per transistor but in this set of conditions it looks like both transistor only add 50pF. That assumption is based on the difference in transition time with change of C3. In any case there is power losses in both transistors during transition due to commutation but these losses are lower than the output transistor power would be without inductor commutation as expected.

• Rg=100R is optimal but can be reduced or increased to 220R makes no significant difference. 

• C3 can be reduced to 220pF before losses start to increase. C3=150pF is usable though. Alternatively, 330pF ensures the lowest power with some margin for component tolerance.

• Reducing the dead-time to just wide enough is not beneficial to 200nS, 500nS works in all cases equally efficiently.

• D3 & D4 included in this simulation and shown in the circuits below might protection components by reducing under or over voltage swing due to Vsd(peak). The diodes also clamp the negative swing of the gate due to Cdg capacitance coupling due to inductor commutation current but in any case, the transistor has protection integrated. With C3=100pF there is a negligible increase in power of about 10mW. 

Transformer output Conclusion

Efficiency is improved but the output voltage range has little margin so turns ratio calculation needs to be with Vs minimum value. That is there will be a large 100Hz AC component output from the PFC function on top of Vs so Vs minimum value needs to be at least 400V.

Reducing R2 to zero ohms and adding a high-value gate turn-on resistor makes the circuit much more efficient.

The transistor losses can be kept modest up to 100KHz normal running. Increasing the value of C3 saves power but adds losses in warm-up mode see below.

D3 & 4 are optional and may as well be included if a dual diode package is fitted. The type shown does not have adequate peak forward current rating though but is a useful place-holder for simulation. BAV99T would be suitable.

Gate driver selection;

Infineon - High side withstands -25V only.

• IR2109s 540nS dead-time, 120/250mA

• IR2111s 650nS dead-time, 250/500mA

• IR2183/4s 500ns dead-time, 1.4/1.8A - Price about $1.40

Texas Instruments - no special protection required.

• UCC20255 TI both drivers isolated and one PWM input. dead-time 8nS (no resistor), 200nS (20K) 4/6A. price about $2.00 ----- This driver does not require the protection against negative voltage in caused by tRR in the commutation current that the IRxxxx requires.

Analogue Devices - not suitable.

• ADuM3223C/ADuM4223C ADI 5V supply and digital inputs also require an inverter to one side input. - not convenient?

• ADuM7223C 5V supply and digital inputs also require an inverter to one side input. - not convenient?

ST - good single driver.

• STCAP2SCM_ ST - This is a single gate driver with negative and positive input options so the top transistor can be driven from the +in and the bottom transistor driven from them -in. Additionally, there is a gate pull-down clamp which will be more efficient than the diode used in the model giving a slightly faster turn-off time and a little better anti-miller capacitance clamping. $1.50 although requires 2 off and other parts this part is included because the PCB layout may be better consequently.

Toshiba - Good single driver, fast for optoisolator type.

• TLP5832 Toshiba - fast single optoisolator, min 5mA in, >1A >1.6A output. 

Infineon Galvanic isolated types. Require logic inverter. SLDON=0 to selects logic supply input regulator and a series resistor must be added. Dead-time setting resistor to be added.

• 2EDF7175F 1A/2A, fixed dead-time, 3.3V logic but there is a regulator requires a series R.

• 2EDF7235K 4A/8A, dead-time control, 3.3V logic but there is a regulator on a chip requires a series R.

• Both of these parts are more complicated but do resolve the under voltage Vsd(peak) protection issue provided a series resistor is fitted with the bootstrap supply.

Conclusion gate driver;

UCC20255 TI - Has been chosen because the most required features are integrated. Few extra parts are needed and the compromises are minimal. That although the part has a higher standby current than IR2184 at 2.5mA instead of 1mA a compared at operating frequency has not been carried out?

The drive has changed from the first circuit so that only pull-down (turning off) current drive is important. The turn-on is intentionally slowed to minimise miller-capacitance turning the otherwise turned off transistor on. Secondly, the turn-on speed is also minimised in order to limit the maximum ID of the transistors.

• Revised because it turns out that slowing the transition speed is also improves efficiency but peak current needs to be re-checked. These changes have been incorporated above.

• The original strategy of fast turn-on and turn-off gate drive would ensure minimum commutation losses in the substrate diode bypassing some of that current more efficiently through the MOSFET but it turns out results in other greater losses elsewhere. This is because the commutation current is much more varied than in a straightforward conventional switch mode power supply so required dead-time, to avoid high or the maximum ID being exceeded during output transitions, changes a lot.

Some of the steps taken earlier but developed further;
Transistors were changed too; IPD60R600P7

Gate charge total; 9nC (5 to 5.5V). note this figure is lower giving lower switching losses but the trade-off is higher Ron (losses).
Internal gate resistance 6.3R.
No maximum gate current is given?

With no extra gate resistance the maximum gate drive typically is;

• Turn on;

• The design has been superseded and crossed out because a high-value gate resistor is used in turn-on case and the gate current is <250mA.

• Turn off; 1.2A = (5.5V - 0.5V) / 6.3R (at the plateau voltage - diode drop)

• 7.5nS = 9nC / 1.2A is roughly the fastest possible switching off time. Other figures on the data sheet give a higher value.

• Absolute transistor turns off occurs at <3V so the fastest turn off time by gate control is higher but still less than <15nS because the total charge occurs at the plateau voltage 5.2V.

When modelled with Rg = diode + 100R or 1K,  1K yielded the lowest power consumption when running normally in circuit AL-0028-01E.

With further work above 100R looked likely to be a better compromise between starting condition but after simulation 220R works well.

• The gate drive simulation it can be seen that the gate voltage rises much more quickly to the threshold voltage just before plateau voltage where the higher Qc occurs. Consequently increasing Rg reduces the power wasted further found by trial and error. Gate capacitance value is given of 363pF, 
• dt=c.dv/i
• 22nS = 363e-12 * 3 / 50e-3 ------ this is fast does not include a delay time if that is relevant but evidently works in the simulation.

• The benefit is small but there is no disadvantage in increasing the gate turn-on series resistor further.  

Gate drive conclusion; This extra arithmetic is necessary in order to understand what happened when simulated and what changes can be made and why they are likely to be beneficial.

Modelling with the output load disconnected to estimate worst-case peak transistor currents starting and warming up;

This simulation gives an indication of maximum current and power during starting and warming up modes. The polarity of the inductor current varies as the operating frequency changes at or near resonate frequency of L3 & C2. The supply voltage is increased to 450V compromises the efficiency a little. This is similar to AL-0028-01? above but with the output inductor disconnected so there is no commutation current available to drive the output transition and transistor have to drive there own losses plus C3 charging current.

Two fast diodes have been added to the model below gate drive in an attempt to turn the MOSFET on more quickly but clamping the voltage to within the driver rated negative swing. I have not seen this done before in the top transistor drive but the bottom transistor is the same as using a gate driver with built in clamp and diode. There is a problem though the reverse polarity MOSFET turn-on characteristic is not specified* - so the idea may be a waste of time trying to model?

* I am advised that the MOSFET characteristic is symmetrical in a recent support request to Infineon. 

AL-0028-02E Transformer output to improve efficiency. 20KHz running and 50KHz starting 

up. This circuit shows the output disconnected but for the snubber capacitor C3. The output 
may behave as an inductive or capacitive load as the frequency is ramped for starting.

This simulation tests the worst case condition, as far as is practical. The transistor maximum ID is not exceeded or miller capacitance cause the transistor that is off to turn on that would cause power wastage. The simulation may not be realistic but is helpful and is the best that can be done.

• Transistor < 70W and peak current 5A at 100KHz (5mS).
• Transistor < 70W and peak current 5A at 100KHz (4mS - 5mS).

The two protection strategies one for the gate driver  D5 + R1 and the additional diode D3 conflicts and could waste considerable power waste in Tr3. In any case, the driver proposed has changed so D5+R1 are not required. THIS CIRCUIT DOES NOT SHOW CHANGES DISCUSSED ABOVE BUT SHOWS THE PROBLEM WITH D5

THE PROBLEM WITH D5 DOES NOT EXIST WITH THIS SIMULATION BECAUSE THE OUTPUT IS DISCONNECTED AS THERE IS NO INDUCTIVE COMPONENT IN THE CIRCUIT.  THEREFORE THE EXTRA PARTS DO NOT INTERFERE WITH THIS SIMULATION RESULT.

The peak current was 16A with Rg=56R but by increasing Rg to 220R reduces the current to 5A peak with C3 = 330pF at any frequency. This is not as expected by calculation.

Modelling transistor selection AL-0028-01? & AL-0028-02? with output connected above;

Although IPU50R950 Transistor was tried the losses were higher at 3-8W but that result is out of context with the present design. The switching time looks like 70nS from the graph. With <50nS looked possible. 200nS looked like the minimum dead-time without losses being increased.

To do later, Consider other transistors such as;

• STD8N60DM2 - ST MDmesh DM2 MOSFETs which are specifically designed for half bridge circuits with commutation currents and claim to have a good low tRR diode integrated.

• IPN60R360P7S - (SOT223) is a smaller part but would have lower Ron figure but unlike parts also considered above has a little higher but still low switching losses.

Continuing to use transistor type; IPD60R600P7

Modelling gate drive and snubber with AL0028-01? & -02?;

Various values of C3 100pF to 330pF, Rg 100R-500R-2K but 1K worked well with zero dead-time gave transistor's running power; 800mW to 2W. This was with an earlier circuit the detail I have removed.

Below are waveforms for the output with the load disconnected. The transition is slow enough not to cause the transistors to turn on due to Miller capacitance. The transition is also slow enough to limit the peak current to below ID max with the gate is pulled low with a high current pull-down.

 AL-0028-02E The output is disconnected. 20KHz and 50KHz This model most likely give 

a more severe than the real starting condition but ether-way there is not much that can be done.
The higher pulse conditions that the transistors can withstand have not been modelled them.

• Red - Transistor power briefly 1.5KW 

• Purple - Transistor current peaks below ID at 7A. The bottom transistor shown's peak is lower than the top transistor's?

• Green - Transistor output voltage is about <100nS wide.

• Blue - gate drive you can see if expended the plateau voltage.

The 7A peak, switching time observed on the graph is 50nS. About 100nS was expected but the arithmetic was a simple approximation. 75uJ = 50nS * 1.5KW as long as this is not maintained long then the transistors can withstand this condition.

 -----

Some of the work on striking and starting the lamp from below has been incorporated here.

Conclusion of Transformer and revised gate drive design;

The operating frequency can be increased but the transistor power loses increase above 100KHz running - there is little befit in doing this other than to use a smaller lower value inductor. The choke selected below looks fine and does theoretically improved efficiency over the original design.

The losses can be reduced by lowering the snubber R to 0R but some resistance may be required in order to minimise EMI (Electromagnetic Interference). 

Although the extra diode around the gate of each transistor was removed the one for the lower transistor could be fitted its effect is neutral but may not be beneficial. The gate driver could be changed to a type that does not need extra diver protection. Although not proven the extra the transistor protection is worth adding because the operation is not modelled with a real PCB parasitic. With the same care, PCB layout should remain as robust as earlier versions have been.

Both the transformer have saved power (50%) and also the revised gate drive (50%). This means that the PCB can be used with a metal area of 60x60mm and it only need be on one side of the board thereby minimising the radiated electrical emissions. 

Some heat-sinking for the starting condition is required and thermistors could be added to sense and shut the power supply down if the power supply does not come out of starting conditions in a reasonable time.

Magnetics

Power Choke - before we finalise the frequency of operation

• Current circulating at various multiples of the running frequency.
• Peak current at 1KV peak (could be over 1.5KV but the circuit will be limited current or voltage monitoring)
• Impedance = 2.pi.f.l
• Current = V/2.pi.f.l
• 20KHz and 50KHz, L = 470uH

• 2.5*f; 7A = 1x10e3 / (2 x 3.14 x 50x10e3 x 470 x 10-6) 
• 14A pk-pk 

• 3.5*f; 5A = 1x10e3 / (2 x 3.1 x 71x10e3 x 470 x 10-6) 
• 10A pk-pk

• 5*f;  3.5A = 1x10e3 / (2 x 3.1 x 100x10e3 x 470x10-6) 
• 7A pk-pk

• 7.5*f;  2.33A = 1x10e3 / (2 x 3.1 x 150x10e3 x 470x10-6) 
• 4.7A pk-pk 

• For 50KHz operation L=220uH and at; 100KHz, L= 100uH.
• 1Arms, 3A = 2x1.4A pk-pk, 

Using Coilcraft web coil-loss modeller;

• It is necessary to enter a small amount of DC current. I set the value to 0.1A the minimum.

• AGP4233-470ME CoilCraft is 42x36x28mm
• Others, MSS1278, MSS1583 are smaller and lower power.

• CoilCraft web-model for; 470uH at 20KHz running, 3A pk-pk.
• MSS1278, 410mW 54'C 
• MSS1583, 760mW 68'C
• AGP4233, 4W 76'C

• Starting 50KHz (f x 2.5) 14A pk-pk,
  
  • MSS1278, 24W
  • MSS1583, 46W
    
  • AGP4233, 160W

• Starting 71KHz (f x 3.5) 10A pk-pk,
• MSS1278, 23W
• MSS1583, 43W
• AGP4233, 98W

• Starting 100KHz (f x 5) 7A pk,
• MSS1278,15W off the graph.
  
• MSS1583, 28W inductance drops to 100uH
  
• AGP4233, 59W inductance drops to 400uH
  

• Starting 150KHz (f x 7.5) 5A pk, 
• MSS1278, 13W inductance drops to 150uH
  
• MSS1583, 13W inductance drops to 250uH
  
• AGP4233, 40W inductance drops to 450uH
  

• CoilCraft web-model for; 220uH at 50KHz running.
• MSS1278, 250mW 37'C
• MSS1583, 450mW 52'C 
• AGP4233, 1.1W 39'C

• Starting 250KHz (f x 5) 7A pk-pk,
• MSS1278, 23W L reduced to 80uH 
• MSS1583, 40W L reduced to 120uH. 
• AGP4233, 66W, L unchanged. 

• Starting 350KHz (f x 7.5) 5A pk-pk,

• MSS1278, 17W inductance reduced to 165uH.
• MSS1583, 31W, inductance reduced to 195uH.
• AGP4233, 44W, inductance unchanged.
  

• CoilCraft web-model for;100uH at 100KHz running. 
• MSS1278, 550mW, 50'C 
• MSS1583, 1W, 125'C
• AGP4233, 1.8W, 50'C

• Starting 250KHz (f x 2.5) 14A pk-pk, 
• MSS1278, 35W off the graph
• MSS1583, 71W off the graph
• AGP4233, 90W, L reduced to 60%, least 'C rise.

• Starting 500KHz (f x 5) 7A pk-pk,
• MSS1278, 36W L reduced to 80uH
• MSS1583, 68W L reduced to 90uH
• AGP4233, 45W, L unchanged

• Starting 750KHz (f x 7.5) 5A pk-pk, 
• MSS1278, 28W inductance reduced to 95uH
• MSS1583, 53W, inductance reduced to 95uH.
• AGP4233, 75W, inductance unchanged.
  

• DMT2-380-2.4L CoilCraft is an input or output filter choke and may not be suitable little is said about whether is it a low Q type. 380uH, 2.4A Torodal, 36 x 36 x 23mm

• DMT3-402-3.7L CoilCaft but as DM2 above. 402uH 3.7A, 41 x 41 x 23mm

MSS1278-474KL_

470 ±10%

707.5

786.2

2.2

1.34

1.54

1.64

0.66

0.90

It is unnecessary to increase the frequency due to the availability of a suitable power choke. Increasing the operating frequency to 100KHz with 500KHz starting looks very good though but the frequency will need to ramp to 750KHz and this may not be possible. This should be simulated.

Transformer;

V-uS at; 20KHz at 450V gives a much higher V.uS value than the transformers available.

• 20KHz; 5.7V.mS = 25uS * 450V / 2

• 50KHz; 2.25V.mS = 10uS * 450V / 2. 

• 100KHz; 1.13V.mS = 5uS * 450V / 2.

Coilcraft, Wurth Electronics, Pulse Electronics all replied and correspond with me on a custom transformer. They can not do any design work without a likely order from an estimate. All three re-looked at off-the-shelf parts but confirmed there that they could not offer anything.

• The major transformer and coil winders require a project and an anticipated purchase quantity in order to proceed to design and make a prototype transformer. Therefore this part of this design cannot be completed.

Coil Craft P3717-AL ---- Line filter common mode choke with high leakage inductance configured as an auto-transformer and also providing the series inductor, looks promising;

• 25mH, leakage inductance 1mH (I will need to check but that could amount too 500uH), 3A, 0.3R, 1,500Vrms. 

• 44.5 x 39.5 x  33mm

• The likely drawback is that inductor has low Q so the resonant voltage may not be high?

Current Transformer;
20KHz, 15V. Gives V.uS; 750 V.uS = 50uS x 15V (would be duty cycle near 100%).

•  Pulse Electronics - some arithmetic required.
• PA10005.xxx - both types will only produce a low voltage. (8.4x7.2x5.5mm)
• PA820xNL  
• 1:125 would only produce 2.7V at 20kHz duty 98%.
• PA1005.125NL Is small and will need extra circuit components. Will provide  (~4V) provided with a minimum operating frequency is 30KHz. 
• The PCB clearance between primary and secondary windings is only >1mm which may not be adequate?
• The PCB clearance between primary and secondary is not adequate for Earth but a connection to Neutral is better anyway. Hipot voltage is 500Vrms but is okay for 220Vrms.

•  Coil Craft
• CST2010-100L_ is only; 254 V.uS
• SCS-100L_ is only; 160 V.uS
• CS4100V-01L is only 298 V.uS
• CST2020-100L is only 395 V.uS 
• 200:1 sense no power
• CST2010-200L_ ; 508 V.uS $1.32 (15x20x10mm)
• SCS-200L_ ; 320 V.uS $2.50 (15x15x10mm)
• CS4200V-01L ; 596 V.uS
• CST2020-200L  791 V.uS
• CST2020-300L  1186 V.uS (300:1)

• Wurth Electronics - none suitable. 
• MID-SNS Sense Transformers (15x20x10mm)
• 1:200 sense only no power
• Part No. 750316796
• 496 V.uS

Controller IC
UC3846 TI - Used originally and is still available. This is still a good choice
UC3856 TI - Has a higher output drive version.
UCC3806 TI - Is a lower power version and is a good candidate?


Striking the lamp - resonance.

I found this quite difficult to model. This is because waveforms change and the power range is so wide. The starting condition is only partly modelled and some high-stress conditions mitigated but not all of those stress conditions modelled are mitigated or likely to occur anyway.

When modelled the power transistors consume a lot of power 150W or 4KW depending on which model of the AL-0028-0?? is used. Although the power is brief the transistors could be very overstressed. The slow turn-on gate drive modification saves a lot of power when running normally but at start up the circuit will be operating at above, on or below resonant frequency and the current waveform is different in each case so this looks like the likely cause of the problem by inspecting the wave-forms.

AL-0028-03D starting simulation 500KHz, 500nS dead-time. 100KHz - 500KHz
Changes; D5 + R1 removed. D1,2,3, and 4 changed to BAV99, Rg = 220R+1R

Simulating at various frequencies is complexly multi-variable even though the model is simplified as much as possible. The model does not include control so the current is not limited the power circulating will drop quickly after the lamp's arc is struck. So the modelling results will give very higher power losses.

• 20KHz running and starting 50KHz was also simulated and works well.

• 100KHz running and 500KHz starting with 500nS dead-time. The power was ~400W per transistor 290mA at 1,500V out. The control will limit the power very quickly after the lamp is stuck <1mS as can be seen in the waveform below.

• 20KHz running and 110KHz starting were modelled and the transistor power was 5W and 330W.

• 20KHz running but starting at 160KHz and 230KHz the transistor power stayed the same but the output current reduced in other words there was no benefit. Above 160KHz there was insufficient current to a resistive load for the current sense to comfortable exceed 150mA adjust the operating frequency during starting. In other words, 10x the running frequency in this case 230KHz does not work.

AL-0028-03D 100-500KHz starting at 500KHz. The lamp should start before the power and the peak current in the transistors becomes excessive. From the graph, the lamp will strike within 16uS.

The simulation does not clearly show what happens. It can be seen that the current in the transistors is <ID max which is the important parameter.

• Switching to a lower or no C3 snubber - carried back from AL-0028-04? below (1mS); 
• C3 = 0pF, 100KHz - 500KHz, 200nS, test 500KHz. Transistor power is; 500W & 2KW.
• C3 = 22pF, 100KHz - 500KHz, 200nS, test 500KHz. Transistor power is; 415W & 460W.
• C3 = 47pF, 100KHz - 500KHz, 200nS, test 500KHz. Transistor power is; 420W & 470W. (<390W 20uS run time)

• Switching C3 down to a lower value for starting and warm up does no harm at starting and may improve efficiency when warming up. 47pF looks optimal without simulating to infinitum.

A further revision to AL-0028-03? to increasing C3 - increased to reduce emissions
The modelling uses excessively high lamp starting current at lower frequencies. This is a very brief behaviour before the controller reduces the lamp current and therefore the total power.

Various combinations of frequency were simulated up to 20KHz, 50KHz with starting frequency 2.5x 5x and 7.5x plus 100KHz, 2.5x and 5x. Capacitor C3;

• 1nF to 2.2nF worked but the dead-time needed to vary.

• For example, it may be possible to adjust the dead-time from the same analogue signal that drives the frequency ramp?

• A disadvantage is the power consumption is higher at starting to see simulation AL-0028-02? 

This option is worth keeping in mind depending on the outcome of EMC testing.

Revise models for normal running.
When the new snubber values, particularly the higher value capacitor C3 is put back in the model the transistor's power consumption rises and it can be seen that depending on the resonant frequency and therefore the value of C3 the dead time has been increased. C1 the coupling capacitor's value also needed to be increased because it's reactance was adversely affecting efficiency.

Increasing the normal operating frequency may allow us to select smaller wound components and give a range of output transformer possibility. It seems from above work that a factor a resonate frequency for starting of 5x normal running frequency works but 10x the circulating current may not be high enough. But at only 2x the current is impractically higher than the running current resulting in potentially unworkable high currents but also the current waveform becomes more like tuned circuit so the current mode controller may not work. These things can be re-tested by changing the parameters in the earlier models above.

   AL-0028-04D 100KHz running, 500KHz starting. 500nS dead-time, 700R load. Warm up 
simulation is highly variable the current direction changes with frequency and load so the 
transistor power can be low or high.

The power dissipated as the frequency passes through sub-multiples of the resonant frequency. The highest figure for each frequency and C3 value gives the best indication of worst-case transistor power.

Arithmetic;

C3 snubber capacitor cannot be increased in value to reduce EMI because power is consumed and seems to interact with resonances. Also calculating an optimum dead-time for the warm-up current also does not work.

• For I=150mA min dead-time dt = c.dv/i; (current limit during starting)
• x7.5 600nS = 100pF
• x5; 400nS = 150E-12 * 400 / 0.15
• x2.5; 1uS for 330pF

• It turns out that this simple assumption is not good enough and a longer dead-time is required or high power consumption and heat-sinking is required at 100KHz running.  

• It also turns out that reducing the running frequency to 50KHz halves the losses to 5W. But increasing the dead-time does not reduce the losses further. At 50KHz running the losses are reduced but increasing the dead-time causes the ramp to invert due to resonance so there is no befit. 

The power in the transistors simulated is up and down and cannot be relied on as a good prediction of what will really happen. Specifically, a heat-sink or the operating frequency reduced is required but may not be necessary ultimately?

• Unless stated C3=150pF, 330pF to 2n2, 500nS and 1uS, Rg=100R. Version -04C with driver protection.

• Running; 100KHz, Start; 500KHz, Test frequency 200KHz, dead-time 500nS, transistor power; 15W.

• Running; 100KHz, Start; 250KHz, Test frequency 200KHz, dead-time 1uS, transistor power; 3W & 800W.

• Running; 50KHz, Start; 250KHz, Test frequency 100KHz, dead-time 1uS, transistor power; 5W. 

Removing the extra gate drive protection may have been beneficial. Reducing C3 is necessary in order to reduce transistor excessive power.

• D5 & R1 removed D3 & D4 remain, Version -04D with driver protection removed, 1 - 2mS;

• Running; 100KHz, Start; 500KHz, Test frequency 200KHz, dead-time 500nS, C1=47nF;
• Rg=220R, C3=150pF, transistor power; 11W.
• Rg=220R, C3=220pF, transistor power; 21W.
• Rg=100R, C3=220pF, transistor power; 26W.
• Rg=220R, C3=330pF, transistor power; 39W. 
• Rg=100R, C3=330pF, transistor power; 48W.

• Running; 50KHz, Start; 250KHz, Test frequency 100KHz, dead-time shown, C1=100n;
• Rg=220R, C3=220pF, 1uS, transistor power; 7.5W. 
• Rg=220R, C3=220pF, 2uS transistor power; 7.5W. 
• Rg=220R, C3=220pF, 500nS transistor power; 11.3W.
• Rg=220R, C3=330pF, 1uS transistor power; 17W
• Rg=220R, C3=330pF, 2uS transistor power; 20W
• Rg=220R, C3=330pF, 500nS transistor power; 25W

• Running; 20KHz, Start; 100KHz, Test frequency 50KHz, dead-time shown, C1=220n;
• Rg=220R, C3=220pF, 500nS transistor power; 245mW.
• Rg=220R, C3=220pF, 1uS transistor power; 250mW.
• Rg=220R, C3=330pF, 2uS transistor power; 1.8W. **
• Rg=220R, C3=330pF, 1uS transistor power; 2.3W. 
• Rg=220R, C3=220pF, 2uS transistor power; 4.5W. **
• Rg=220R, C3=330pF, 500nS transistor power; 10.5W. 

AL-0028-04D 100KHz operating 500KHz starting but 500nS dead-time. High 

power pulse shown in blue and red is not eliminated when the dead-time is increased 
to 1uS because the voltage does not continue to fall but may curve and rise.

The output drive is one of the Red waveforms - at different frequencies, it can curve back or complete the transition within the dead-time. So the power consumed in the power transistors can vary widely. As discussed above but a solution may be to switch in and out an additional C3 depending on the mode warm up or running.

• C3 = 0pF, 100KHz - 500KHz, 500nS, test 250KHz. Transistor power is; 12W.
• C3 = 22pF, 100KHz - 500KHz, 500nS, test 250KHz. Transistor power is; 8W.
• C3 = 47pF, 100KHz - 500KHz, 500nS, test 250KHz. Transistor power is; 4W.
• C3 = 47pF, 100KHz - 500KHz, 200nS, test 250KHz. Transistor power is; 2W.
• C3 = 0pF, 100KHz - 500KHz, 200nS, test 250KHz. Transistor power is; 1.6W.
• C3 = 22pF, 100KHz - 500KHz, 200nS, test 250KHz. Transistor power is; 1W.

A heat-sink would be a safe option, this option works to some degree. There is a pattern anticipated that the shorter dead-time works better because the output transition is completed whilst the higher harmonics of the transition predominate.

Conclusion of warm-up mode;

The gate drive protection is removed and C3 is reduced in value, excessive power in one transistor now does not occur some times. The power in transistors still goes up and down in an unpredictable way with parameter changes. Using 50KHz starting at 250KHz design looks good. Trying 100KHz starting 500KHz could be prototyped.

Put back extra diode D3 & D4; Probably won't help the turn-on time would still be <100nS = 60nS + 5nS +?. There is no harm in fitting those diodes.

Current control - simple simulation basic operation;

Reconsidering current feedback with the lamp as a resistive load and the transformer output

 AL-0028-05A 100KHz running for 500KHz starting. R4 is the lamp. It can be seen that in this 
case the current measures 8% to 12% low so increasing R2 to 22R would correct the error. 

• Simulation with  10R, 100KHz running for 250KHz starting.
• V_RMS; 9.9V is 990mA but mean measured is 855mV or 855mA
• Lamp current in resistive load; -1.4A & 1.8A peak

• Simulation with  87R, 100KHz running for 250KHz starting.
• V_RMS; 87V is 875mA but mean measured is 833mV or 833mA
•  Lamp current in resistive load; -1.25A & 1.25A peak

• Simulation with  10R, 100KHz running for 500KHz starting.
• V_RMS; 9.95V is 995mA but mean measured is 878mV or 879mA
• Lamp current in resistive load; -1.5A & 1.9A peak

• Simulation with 75R, 100KHz running for 500KHz starting. 1:1
• V_RMS; 75.0V is 1.0A but the mean measured is 919mV or 919mA
• Lamp current in resistive load; -1.25A & 1.25A peak

It can be seen that the current feedback can be recalibrated and will then fit a range of lamps well. With resonance set at 250KHz the output voltage is higher with a high value resistor load.

It was observed above that for resonance set at x2.5 the running frequency that current does not drop as the frequency is reduced whereas at x5 the running frequency the current does drop then rise again as frequency was reduced. The lamp power supply works both ways and the lamp starting control was stable in practice both ways.

The Controller; 

The controller below included current mode switch-mode power supply IC, the Starting and warm up function and the Snubber switch drive called efficiency. Developing this part of the circuit is beyond the scope of this project. I have included a design to indicate without using more text how other functions may work.

 AL-0028-06A4 CADSTAR 2018 version. The ramp generator comparator is an op-amp used in comparator 

mode the circuit has extra filtering around this circuit in order to minimise the interference by EMI on the 
circuit. Function bottom left IC2-B. Bottom Right Gate driver using discrete semiconductors.

The switch mode power supply oscillator frequency has been set to 50kHz and starting up < 400KHz using component values taken from a graph in the data-sheet. The shut-down pin comparator protects the circuit from over voltage with C11 proving a retry soft-start delay.

The differential feedback input is useful because the voltage is negative. The offsetting resistor network is included to make all voltages positive necessary for the controller IC's inputs.

This controller's maximum operating frequency is 1MHz so the maximum switching frequency is 500kHz (50%). It may be necessary to select another controller type or reduce the desired starting frequency. Below 30kHz and 150kHz has been chosen so different oscillator components are required.

C4 is included to overcompensate the op-amp but it's value needs to be checked. The capacitor may not be required if the function block is placed on the other side of the PCB to the magnetics. Therefore to keep the op-amp stable in this likely high EMI environment.

• The ramping frequency is set quite low which does not matter much but that frequency could be increased to say 20Hz. (From simulation above the tuned circuit reaches a high enough voltage in a few milliseconds so the frequency sweep can be faster).

Half-bridge output and arc-lamp;

The circuit has a number of undefined parts so some of what is shown are place-holders. The arc-lamp supply could be connected to Earth but it is connected to Neutral instead so that noise or earth fault does not affect the over-voltage sensing. Therefore the causes of faults or malfunctions will be simpler to determine either the lamp unit or the mains power but unlikely to be a combination of those.

 AL-0028-06D (CADSTAR 18) Half-bridge with switched snubber capacitor for efficiency. 

Output of the output transformer referenced to Neutral (but not to Earth)

The Current Transformer is tested to 500Vac but is not specified for mains Earth isolation. In practice 3mm, 6mm or 10mm clearance is only applied to the input filtering before the fuse and to Earth everywhere. The distance between Neutral and live circuit is closer after the fuse. This is how I observe the standards are applied.

A thermistor is placed near the top power transistor to shut the power supply down if the transistor gets hot. This should occur if the power supply does not move on to the full current stage but stays in the warm up mode.

VSSB and C21 (0V) could be connected to the low transistor S (Isens01) this would be the conventional connection with this driver. There is probably a marginal difference in this case and I have not tested that option when selecting a different driver to the first one chosen. At some point, the design cycle has to stop though.

Reading standards does not answer some ambiguities which the industry that uses and wrote those standards understands. Getting advice is a good strategy even though you (the company you work for) are responsible for your design and manufacturing.

Outstanding issues;

• Determine a suitable output transformer OR;
• Consider an output transformer with suitable leakage inductance.

• The operating frequency chosen is 30kHz and starting frequency 150kHz.
• Alternatively operating frequencies; 50KHz or 100KHz have been modelled. 
• Previous version 100kHz half-bridge uses smaller SMT parts; 100kHz option half-bridge (-06B) 

• The tuning capacitor and inductor voltage ratings are too low. They need to withstand 2kV end to end and to the casing.

• Design PCB with metal heat-sink or copper area on the PCB. 
• Both options are provided.
• Thermal reliefs have been turned off on the planes.

Half-bridge section part PCB layout design evaluation breadboard AL-0028-06C High-frequency designs 

if not all designs are best carried out on a PCB in order to achieve a realistic evaluation and refinement.

Both copper PCB area for heat-sink and mounting for metal thermal mass has been provided for although both are not required ultimately.

• 0V planes and heat-sink both sides to have vias added ~ 10mm. Tracks routing and spacing to be refined.

• Power choke L7; AGP4233 330uH output choke. But smaller foot MSS1278 or MSS1583 but operating at up to 100kHz could be tried.

• T1 has not been sourced yet. Voltage isolation will need to be checked. Therefore a row of four pins is a place holder on the PCB for that undefined part.

• C16, a suitably high enough AC voltage rated part has not been found but a 3n3F 2kV polypropylene is placed. 1nF 2kV COG for 500kHz starting and other options have not been placed. 

• Current sense transformer has been chosen.
• The clearance on the PCB is only >1mm may not be adequate for safety L to N although the voltage is 500Vrms will work.

• The clearances are to comply with functional safety level so in addition to checking Live to any other point on the circuit it looks like the power transistor pads need to be altered for example?

Revised low voltage controller power supply;

AL-0028-06C Low voltage controller power supply uses power from the PFC.
A protection thermistor is placed close to the top power transistor on the half-bridge sheet.

The minimum operating frequency is 30KHz. This uses a larger current transformer but fewer components. Tr5 powers the controller.'s power, initially C14 turns on the controller's power. It does the same as version -06B which has more components but a smaller current transformer. R11 is added to ensure that the power supply clearly shuts down or runs rather than hovers trying and giving up just on the shutdown threshold.

Conclusion of current sense and low voltage PSU;

• Set the operating current to 30KHz and starting at 150KHz
• Therefore the inductor is changed to 330uH type; AGP4233.
• A metal plate heat-sink option is included to try first.

Conclusion for the revised arc-lamp function;

In conducted emission tests the initial circuit had a much cleaner EMI profile than a traditional main supplied inductor and starter. With a traditional mains supplied ballast it can be seen that the lamp re-strikes on every zero current crossings which presumably will also increase the number of other colours of light emitted as well as making the light flicker at 100Hz. The high-frequency lamp with the large DC reservoir capacitor produces a very clean uniform light.

A real circuit will need to be made and some of the various starting up frequencies tested for emissions and efficiency these cannot be modelled or cannot be modelled very accurately.

It is possible for this design to comply to a higher standard than basic CE marking thereby making the arc-lamp unit particularly useful for sensitive scientific measurement.

--------------------------------------------------------------------------------------------------

 Power Factor Correction; fly-back converter

• IC type; ucc28056 TI looks interesting low start-up current few pins and low component count. I suspect that the inductor must not conduct continually even at maximum load so that there is a no current period for the double function input pin can sense the input voltage. Such operation is likely to cause EMI but it is worth reading the data-sheet and considering the input filtering despite what I have said the part will work and should meet all international standards without undue cost.

Interpreting a function diagram in a data sheet tells us a lot. Then reading the text is important of cause to understand detail voltage rating mode of operation.

• IC type; L6562AT ST, is a conventional continuous mode PFC controller but uses a few more resistors. So it may not be such cheap a solution?

I observe that strategy changes in electronics reasoned then changes back again. I mention continuous mode fly-back converter for PFC having lower emission because the inductor tends to ring when it is turned off.  But the very new IC that TI is promoting now (2018) there is a trade-off safe conventional design or potential small cost saving in a future design is a good strategy. Because the design has new risks with the output stage I am inclined to recommend not take a further risk with the PFC at a design meeting with perhaps just 10-20p of saving.

-----------------------------------------------------------------------------------------------------------------------

Conclusion

The analogue design with discrete parts that need optimisation does not stop until you draw a line and say this is good enough now. All circuit design used to be art like this but so much design can be done by choosing function blocks and putting them together. If possible use function blocks to make your life easy, except when they do not do the job well or at a suitable price.

• The original working design proves to be a very useful source of low mains flicker and light fluctuation laboratory light source. This means that the light source can be used with electronic equipment more easily. 

• This arc-lamp power supply weighs a lot less than a conventional mains choke unit. It is unknown if the unit will be very power efficient it is an untested re-design. It is likely though that the power inductor will run much cooler than the ones used decades ago because wound component makers have got much better at stating what their catalogue parts will do.

• There are other ways of correcting light wavelength without using an arc lamp. These other ways may be cheaper but not necessarily easier but even so, this arc-lamp power supply may have too few applications, if any, to be viable commercially?

• This re-design exercise may have no practical application but the exercise shows that there are areas of electronics design that still involve, are art and can not be just put them together using bought function blocks.

• Design with a margin but you should expect function blocks to perform better than worse or typical case calculated. Manufacture of modules, ICs and discrete parts claim. It is rare to find an electronic part that is not understated but that also occurs. A fully integrated Switch Mode Power supply IC's usually claim maximum power rather than realistic practicable power - they will of cause do what is claimed but at room temperature with the part running at maximum temperature.

When it is decided to stop the next step is design review and to pick up the details not picked up earlier. This is where a sledgehammer approach to just get a working design won't do either the design should not only work and meet CE marking as was always the case but should not waste power and this is where this design cycle may be necessary. The circuit is unavoidably unconventional, in any case, needs more work to complete it. The best thing is to stop giving yourself a break then come back and finish off without leaving the last flawed experiments in the design.