SG3525A Plasma Speaker - Project

[davekni]

I will order some higher voltage gate driver ICs to test out (thanks for the link to your post).

Another option to consider:  Run SG3525A on 10V or so and use FET output stage of UD2.7 driver.  Four FETs likely cost less than a driver chip.

[ZakW]

I built the other circuit, I need to repair the bridge so I can take a video. It seems more stable but when I adjust the duty cycle so the audio starts modulating the input the arc starts making a loud hiss. If I pause the audio the hiss is still there and very noticeable. 

Like you said, if I turn the gain down low and the volume up the arc is really loud and for the most part very clear. Sounds better as I turn the input volume up on my phone, however, the hissing remains.

I am using a small Bluetooth module (HiLetgo M18 and m28) for the audio. I used the m28 on my first board which also had a hiss but by adding a 2.2k resistor to the output of the module it was pretty much gone. I did the same with the m18 module but the hiss remains. I can try ruling out the module by disconnecting it and adjusting the duty cycle. I am not convinced it is the module though, any ideas what could be causing the noise?

When the duty cycle is set to max via the POT the arc is completely silent. It is only when I reduce the duty cycle so the audio can be heard that the hiss starts to occur.


Update:

- I removed power from the Bluetooth module and the hiss persisted.

- Now that I replaced the IGBTs I will try and work on a video to demonstrate what is happening.

[davekni]

- I removed power from the Bluetooth module and the hiss persisted.

That is a bit surprising.  Could you try shorting the audio input to make certain there is no voltage attempting to modulate duty cycle?

Now that I replaced the IGBTs I will try and work on a video to demonstrate what is happening.

If shorted audio input still results in hissing, the most helpful information would be scope captures.  See if duty cycle is stable or has jitter.  Same with frequency.  A capture video as in your reply #10 would be great.

[ZakW]

That is a bit surprising.  Could you try shorting the audio input to make certain there is no voltage attempting to modulate duty cycle?

Shorting the input helped a ton, almost became silent. When I zoom in I can still see the duty cycle modulating a little even while shorted. There is definitely noise being caused by something, I suspect the protoboard to be the issue. Especially since the module is all the way across the board and a single wire is running over everything to get to the SG3525.

Overall, this seems to be a much better setup. I expect that if I made a proper PCB the noise issue would go away. The audio is incredibly loud when tuned correctly, almost too loud.

I would like to try modifying my other board by adding the IXDN604PI gate driver IC in for increased gate voltage as well as utilize both SG3525 outputs. I wonder how that might impact reliability and performance.

I should have a video posted tomorrow or Saturday. 

Here are some scope captures of the new build:

Yellow = Vge | This is with duty cycle at max. Longest arcs, almost silent, audio modulation does not work at this setting.


Yellow = Vge | This is with the duty cycle adjusted a bit lower. At this level you can see the duty cycle modulate in response to the audio. Arc is loudest when stretched the longest.


IGBTs are staying nice and cool for short runs. More intensive runs where I am playing loud audio and adjusting everything causes them to get pretty warm.

[davekni]

Now that I replaced the IGBTs I will try and work on a video to demonstrate what is happening.

Vge looks great.  18V is plenty for IGBTs especially since you don't have high current bursts as with a DRSSTC.

This is with duty cycle at max. Longest arcs, almost silent, audio modulation does not work at this setting.

Exactly as expected.  At max duty cycle there is no margin left to modulate duty cycle.

Yellow = Vge | This is with the duty cycle adjusted a bit lower. At this level you can see the duty cycle modulate in response to the audio. Arc is loudest when stretched the longest.

Also exactly as expected.

More intensive runs where I am playing loud audio and adjusting everything causes them to get pretty warm.

Wouldn't expect loud audio to change IGBT heating significantly.  Average duty cycle should remain the same.  Perhaps there is some other condition such as a harmonic of oscillator frequency hitting a transformer resonance.

The audio is incredibly loud when tuned correctly, almost too loud.

By "too loud" do you mean it gets distorted?  Or does it sound fine to someone standing farther away where volume is pleasant level?  Some distortion is expected as audio modulates duty cycle too close to 0%.  Even 10% to 50% may have some distortion depending on how arc power translates to sound pressure.

I should have a video posted tomorrow or Saturday.

Scope video isn't needed if it continues to work well.  My thought was to see if hiss was due to duty cycle modulation or to frequency modulation.  Since grounded input fixes most of hiss, seems likely there is some current noise on input.  That could be from wiring issues or perhaps SG3525A chip input has white noise current.  Noise would not be critical for normal SMPS use.  You could add an opamp buffer or emitter-follower buffer from audio input to SG3525A so that it is driven by a low impedance to minimize effect of noise current (minimize hiss).

[ZakW]

Vge looks great.  18V is plenty for IGBTs especially since you don't have high current bursts as with a DRSSTC.

Thanks, I was happy with it too! I added some additional bypass capacitance and it cleaned up a lot. Since you mentioned a 20v IC I was thinking to drive it at 20v. I will back it down a bit.

Wouldn't expect loud audio to change IGBT heating significantly.  Average duty cycle should remain the same.  Perhaps there is some other condition such as a harmonic of oscillator frequency hitting a transformer resonance.

That is probably it. Just changing a lot of parameters it is hard to keep track and then I do a heat check and things are hot. I chalked that up to adjusting a lot of settings.

By "too loud" do you mean it gets distorted?  Or does it sound fine to someone standing farther away where volume is pleasant level?  Some distortion is expected as audio modulates duty cycle too close to 0%.  Even 10% to 50% may have some distortion depending on how arc power translates to sound pressure.

Too loud as in I wouldn't play my music that loud from my phone  Louder audio seems to sound better, I think it overcomes the noise but also sounds richer due to the increased duty cycle modulation.

Noise would not be critical for normal SMPS use.  You could add an opamp buffer or emitter-follower buffer from audio input to SG3525A so that it is driven by a low impedance to minimize effect of noise current (minimize hiss).

Fair point, the emitter-follower buffer seems simple enough I might be able to lash it to my board to test it out. Thanks for the insight!


Update 1:

Duty Cycle based driver (new):

- I tested the emitter follower, did not seem to want to work with my setup. I followed a generic schematic for it from this video https://youtu.be/1wMqNRvMUsY?si=eXeIIV3z1wYp-VCU&t=301. Duty cycle adjustment became a lot more sensitive, finding the correct combination between gain and duty cycle was difficult. The noise was not reduced.
- I am hopeful that a good low noise layout via a proper PCB will reduce if not fix the noise issues.
- Produces much louder audio with little distortion, minus the hiss.

 
Frequency based driver (old):

- Modified the board so each SG3525 output was going to each input on the IXDN604PI gate driver IC.
- Aimed for 18v Vge
- Seemed a lot more stable, I was able to adjust the frequency more and get a lot closer to resonance before it stated acting weird. This allowed me to reduce the input audio volume which helped to reduce distortion. Sounded about the same compared to only using one output from the SG3525.
- The half bridge gets HOT! The bridge DC supply caps and the copper tape itself get very warm, so did the IGBTs. The other driver does not have this issue.


Update 2:

I recreated the schematic in Kicad. I plan on using the 5v Reference from the SG3525A (pin 16) to provide the 5v power for the blue tooth module. Per the datasheet, Vref can provide 50mA. The module only requires 20mA.

[davekni]

- I tested the emitter follower, did not seem to want to work with my setup. I followed a generic schematic for it from this video

Duty cycle adjustment became a lot more sensitive, finding the correct combination between gain and duty cycle was difficult. The noise was not reduced.

Did you place the emitter follower circuit directly on the audio input before C2?  If so, it should not affect duty cycle.  C2 blocks the DC component of emitter follower output, as is necessary for audio.

- The half bridge gets HOT! The bridge DC supply caps and the copper tape itself get very warm, so did the IGBTs. The other driver does not have this issue.

Not sure why the heat.  Perhaps you are running close to a transformer series resonant frequency (leakage inductance and output winding capacitance).  However, earlier scope captures suggested you were sufficiently under series resonance.  Other thought is switching losses.  Perhaps there is a layout difference that makes this board oscillate at high frequency briefly during each half-bridge transition.  Or some other such artifact.  (I have had issues myself when driving fast FETs through GDTs, where GDT winding capacitance causes unwanted high frequency feedback.)

I recreated the schematic in Kicad. I plan on using the 5v Reference from the SG3525A (pin 16) to provide the 5v power for the blue tooth module. Per the datasheet, Vref can provide 50mA. The module only requires 20mA.

Looks good.  20mA will add power dissipation (heat) to SG3525a chip.  At 20V supply, drop to 5V is 15V, so 300mW of additional heat in SG3525a.

[ZakW]

Did you place the emitter follower circuit directly on the audio input before C2?  If so, it should not affect duty cycle.  C2 blocks the DC component of emitter follower output, as is necessary for audio.

That was my mistake then. I bypassed c2 thinking the 1.7uF cap in the buffer circuit was already blocking DC. That explains why I saw an offset then. I still have it constructed, won't take but a minute to try it out. If it helps I will add it to my PCB.

Update: I tried it and it still did not work, I will keep messing with it to see if I can get better results.

Not sure why the heat.  Perhaps you are running close to a transformer series resonant frequency (leakage inductance and output winding capacitance).

I think this is why. Just to clarify, the frequency modulated driver is the one that causes the heating problems not the new duty cycle one. After updating the driver to use both SG3525 outputs and the new driver IC the added stability of the circuit allowed me to tune a lot closer to resonance (driver output looked a lot more sinusoidal) and that is when things start to get pretty hot.

Looks good.  20mA will add power dissipation (heat) to SG3525a chip.  At 20V supply, drop to 5V is 15V, so 300mW of additional heat in SG3525a.

I ended up removing that and just adding a separate 5v regulator.


How does this look? I am really happy with how it turned out. I was able to borrow a lot from my previous SSTC design for the bridge connections. Dimensions are 87mmx99mm












PCB design question - is it okay to have ICs or other components near the GDT? I tried to provide as much space as I could so as not to induce any signals from the GDT. I am not entirely sure that is a real concern though.

[davekni]

That was my mistake then. I bypassed c2 thinking the 1.7uF cap in the buffer circuit was already blocking DC. That explains why I saw an offset then. I still have it constructed, won't take but a minute to try it out. If it helps I will add it to my PCB.

Update: I tried it and it still did not work, I will keep messing with it to see if I can get better results.

Input cap in emitter follower circuit isolates DC level of input audio from the half-supply DC level of BJT base.  Output of emitter follower is one Vbe (~0.7V) lower.  Needs C2 to isolate that from DC level set by RV1 duty cycle POT.
No idea why it wouldn't work with both caps in place.  For loud audio, the 10k emitter resistor may need to be a bit lower to provide enough drive current.  Depends on what supply voltage you are using for emitter follower.

How does this look?

Looks good in general.  One question:  What is the purpose of R1 in series with audio input?  A series resistor on external inputs can be useful to reduce sensitivity to ESD.  However, this will make audio input impedance 2k higher.  If there is significant noise current on SG3525A input, that 2k resistor will increase hiss.

PCB design question - is it okay to have ICs or other components near the GDT? I tried to provide as much space as I could so as not to induce any signals from the GDT. I am not entirely sure that is a real concern though.

No need for space that I can think of.  GDT looks to be a well wound toroid.  It should have little external leakage magnetic flux.

[ZakW]

Input cap in emitter follower circuit isolates DC level of input audio from the half-supply DC level of BJT base.  Output of emitter follower is one Vbe (~0.7V) lower.  Needs C2 to isolate that from DC level set by RV1 duty cycle POT.
No idea why it wouldn't work with both caps in place.  For loud audio, the 10k emitter resistor may need to be a bit lower to provide enough drive current.  Depends on what supply voltage you are using for emitter follower.

I will try lowered the 10k resistor value. I am using 5v.

Looks good in general.  One question:  What is the purpose of R1 in series with audio input?  A series resistor on external inputs can be useful to reduce sensitivity to ESD.  However, this will make audio input impedance 2k higher.  If there is significant noise current on SG3525A input, that 2k resistor will increase hiss.

Thanks - I saw a video of someone else using a Blue Tooth module for their audio source. They also had an issue with hiss and were able to reduce it significantly with a resistor on the output of the module. They attributed it to "better matched impedance". I too was able to reduce the hiss in my first version by using a 2.2k resistor on the output of the module. I assumed that would also be the case for the smaller module I used in the duty cycle version. I decided to include incase it was needed, easy to bridge the pads if it is not necessary. I can test it out and see if bypassing it improves things, if it does I can remove it from the PCB. 

No need for space that I can think of.  GDT looks to be a well wound toroid.  It should have little external leakage magnetic flux.

Thanks for confirming that.

[davekni]

Thanks - I saw a video of someone else using a Blue Tooth module for their audio source. They also had an issue with hiss and were able to reduce it significantly with a resistor on the output of the module. They attributed it to "better matched impedance"

A resistor helping makes sense, until you said the hiss continued with Blue Tooth module powered off.  However, I don't think impedance matching is the issue.  Rather many (probably most) audio outputs include some high-frequency energy.  Too high frequency for humans to hear (and for most speakers or ear phones to convert to sound).  However, that high frequency content will alias with your SG3525A frequency creating low frequency noise (hiss).  A series resistor combined with input capacitance of whatever is after the resistor makes a crude low-pass filter.  A series resistor followed by cap to ground would make a better low-pass filter.  Much fancier low pass filters are possible.
BTW, the high frequency noise may be from two sources.  For low-level audio outputs (outputs not intended to directly drive headphones or speakers), high frequencies come from the sigma-delta audio DAC used.  For higher power outputs intended for headphones or speakers, many audio amplifiers are class-D, modulating duty cycle of a chop waveform.  That is very much like what you are doing with your plasma speaker, but at a high voltage.

[ZakW]

A resistor helping makes sense, until you said the hiss continued with Blue Tooth module powered off.  However, I don't think impedance matching is the issue.  Rather many (probably most) audio outputs include some high-frequency energy.  Too high frequency for humans to hear (and for most speakers or ear phones to convert to sound).  However, that high frequency content will alias with your SG3525A frequency creating low frequency noise (hiss).  A series resistor combined with input capacitance of whatever is after the resistor makes a crude low-pass filter.  A series resistor followed by cap to ground would make a better low-pass filter.  Much fancier low pass filters are possible.
BTW, the high frequency noise may be from two sources.  For low-level audio outputs (outputs not intended to directly drive headphones or speakers), high frequencies come from the sigma-delta audio DAC used.  For higher power outputs intended for headphones or speakers, many audio amplifiers are class-D, modulating duty cycle of a chop waveform.  That is very much like what you are doing with your plasma speaker, but at a high voltage.

That makes sense, thanks for the info.

What about floating pins on the SG3525? I am not using pins 16, 10, 3 (SYNC), 4(OSC OUTPUT). I think 10 and 16 are fine floating. I tried using 10k pull down resistors on pins 3 and 4, it might have reduced the noise but it is hard to tell.

I really think a lot of this noise is due to ground loops and EMI/RFI from the board. I built it in a hurry and have a lot of wires crossing each other under the board. At this point the noise sounds more like a high pitch tone that is very obvious when I pause the audio but goes away if I adjust the duty cycle to max (arc becomes silent). So something must be modulating the duty cycle when it is adjusted to a certain point to be able to hear audio.

[davekni]

What about floating pins on the SG3525? I am not using pins 16, 10, 3 (SYNC), 4(OSC OUTPUT). I think 10 and 16 are fine floating. I tried using 10k pull down resistors on pins 3 and 4, it might have reduced the noise but it is hard to tell.

Best to have a bypass capacitor on pin 16 even if not used outside.  Lowers noise on reference inside SG3525.  Also best to use pin 16 for 5V to RV1.  Less noise than using 12V, and pin 16 reference matches internal ramp generator.
Pin 10 (shutdown) should be grounded.  Review data sheet for such pins.  It explicitly says not to leave floating to avoid noise pickup causing unintended shutdown.
Pin 3 (sync) looks OK to float or tie to ground.
Pin 4 is an output, so floating is fine.

At this point the noise sounds more like a high pitch tone that is very obvious when I pause the audio but goes away if I adjust the duty cycle to max (arc becomes silent).

Adjusting duty cycle to max (clipped to ~48% by SG3525 internal circuitry) is setting input request beyond 48%.  Input (unless very loud) is just changing the requested duty cycle, say from 80% +-20% so 60% to 100%.  SG3525 still clips to 48% internally.  So of course it is completely insensitive to inputs, both audio and noise.  This experiment you keep repeating is just verifying that noise source is on input half of SG3525 or circuitry before that.  Noise is not inherent arc instability or transformer corona or any other output side cause.
Your description of high pitch noise fits with aliasing of audio source input's even higher frequency components with oscillator frequency.  I'd guess you need a low pass filter on input.  Scoping audio input may help.  May need to set scope gain high (low volts/div) to see audio HF components.

[ZakW]

Best to have a bypass capacitor on pin 16 even if not used outside.  Lowers noise on reference inside SG3525.  Also best to use pin 16 for 5V to RV1.  Less noise than using 12V, and pin 16 reference matches internal ramp generator.

I tried this but duty cycle would not reach max (~48%), it was much lower. When I switched back over to the 12v supply it was fine. I might need to lower the 10K resistor if I use the 5v supply.

Pin 10 (shutdown) should be grounded.  Review data sheet for such pins.  It explicitly says not to leave floating to avoid noise pickup causing unintended shutdown.

Done, didn't help. I went back and looked in the datasheet again, I see where it says to ground it to prevent shutdowns, thanks for pointing that out.

Adjusting duty cycle to max (clipped to ~48% by SG3525 internal circuitry) is setting input request beyond 48%.  Input (unless very loud) is just changing the requested duty cycle, say from 80% +-20% so 60% to 100%.  SG3525 still clips to 48% internally.  So of course it is completely insensitive to inputs, both audio and noise.  This experiment you keep repeating is just verifying that noise source is on input half of SG3525 or circuitry before that.  Noise is not inherent arc instability or transformer corona or any other output side cause.
Your description of high pitch noise fits with aliasing of audio source input's even higher frequency components with oscillator frequency.  I'd guess you need a low pass filter on input.  Scoping audio input may help.  May need to set scope gain high (low volts/div) to see audio HF components.

I tried using a low pass filter on the input. I used a 20k POT and a 6.8nF ceramic cap. Set my function generator to 20kHz to tune to cut off anything above that frequency as I measured the output with my scope. The high pitch noise/hiss was still present, even when adjusting the filter to cut off way lower it did not make a difference.

My grounding definitely has some loops and from a quick search on google that can manifest as a hiss in audio equipment. If I go forward with ordering my PCB and the hiss is not present I guess that would be one way to tell. Not sure what else to try at this point.


Update:

- Recorded a video of the issue. Yellow = Vge.
- Disconnected the module from power and removed the capcitor connection to pin 2. The only thing connected to pin 2 is the POT for duty cycle.
- When adjusting it you can see the signal start to jitter, I believe that is the noise that I am hearing.
- Bridge is not running just the driver.

[davekni]

I tried this but duty cycle would not reach max (~48%), it was much lower. When I switched back over to the 12v supply it was fine. I might need to lower the 10K resistor if I use the 5v supply.

Something is wrong.  Input voltage required for max duty cycle is specified at 3.3V typical, 3.6V max.  5V should be plenty.  If gain is set above 1, even less than 3.3V should result in max duty cycle.  (I'd recommend minimum 1.0 gain setting, adjusting audio input volume instead.)

Update:

- Recorded a video of the issue. Yellow = Vge.
- Disconnected the module from power and removed the capcitor connection to pin 2. The only thing connected to pin 2 is the POT for duty cycle.
- When adjusting it you can see the signal start to jitter, I believe that is the noise that I am hearing.
- Bridge is not running just the driver.

Yes, does look like the hiss issue.  You could repeat this experiment with a 0.1uF or larger cap from pin 2 to ground.  And run at minimum gain in case above was at any higher gain.

I noticed that SG3525A data sheet test circuit has a 10nF cap from pin 9 to ground.  That lowers bandwidth to 10kHz at minimum gain.  (Lower frequency response at higher gain.)  Adding this cap may improve noise.

[ZakW]

Something is wrong.  Input voltage required for max duty cycle is specified at 3.3V typical, 3.6V max.  5V should be plenty.  If gain is set above 1, even less than 3.3V should result in max duty cycle.  (I'd recommend minimum 1.0 gain setting, adjusting audio input volume instead.)

Strange, I will have to measure the output on pin 16 to verify it is at 5v. I can also check pin 2 while I adjust RV1.

Yes, does look like the hiss issue.  You could repeat this experiment with a 0.1uF or larger cap from pin 2 to ground.  And run at minimum gain in case above was at any higher gain.

I noticed that SG3525A data sheet test circuit has a 10nF cap from pin 9 to ground.  That lowers bandwidth to 10kHz at minimum gain.  (Lower frequency response at higher gain.)  Adding this cap may improve noise.

I will give both of these a try, thanks for the ideas!

[davekni]

I can also check pin 2 while I adjust RV1.

Also check pin 9, voltage after internal opamp gain.  Should match pin 2 if gain is set to minimum of 1.0.

[ZakW]

I noticed that SG3525A data sheet test circuit has a 10nF cap from pin 9 to ground.  That lowers bandwidth to 10kHz at minimum gain.  (Lower frequency response at higher gain.)  Adding this cap may improve noise.

a 10nF cap from pin 9 to ground did help to reduce some of the noise but not all, out of everything I tested recently that change had the most impact to reducing noise.

Yes, does look like the hiss issue.  You could repeat this experiment with a 0.1uF or larger cap from pin 2 to ground.  And run at minimum gain in case above was at any higher gain.

This also helped a bit to cut down on the noise but was not something I could implement with audio as well.

Overall, ignoring the noise and tuning it the best I can the audio just doesn't sound nearly as clean as the frequency modulation method. Since my goal is to make the best sounding plasma speaker I am going to switch back to the frequency modulation design. If I were to make a general flyback driver, the duty cycle variant is ideal.

Here is the updated PCB:
- Added a power on LED
- I changed a few bypass cap values.
- Added some 100uF electrolytic caps, one to each 12v and 18v supply rail.
- Also added some scope test points to make it easier to hook my scope up to the board.