I came across articles by burson and Benchmark why they have switched from a normal LPS to SMPS. They got unique reasons - hums in the mains operating frequency with lps, impedance reduction/power transients, etc and if I look intuitively I could see things like inrush etc.
Max Current Power Supply – Burson Audio
Audio Myth - "Switching Power Supplies are Noisy" - Benchmark Media Systems
Marketing speak aside, the burson sounds very good and the whole amp is under 200$. I was curious how I could go about replicating this design. I understand that this is kind of an IC/PCB that might need more precision, but I'd love to know the engineering that goes behind all these, and if possible, diy options (atleast a design simulation).
What I see common is that they use really high switching frequencies and I'm not sure how they do it exactly. Burson has a diagram in its page (the rectifier diagram looks wrong lol), but I am unsure what that MCPS would have inside. Asking to a few other engineers they also told that high switching frequencies would have noise other than just switching noise, like electromagnetic radiation etc. and some really complicated compensations in grounding etc are needed in the PCB.
Would love guidance on what are all the things to look around and how to design such an smps? I do understand basics of ac-dc conversion, but only from a utility standpoint, not from a fidelity standpoint (at best I only know about ripples).
On a side note, I'd also love to know what is happening at each stage on the burson fun schematic. I can understand the ac to dc schematic but I'm unable to understand how the first noisy dc is converted into pulses, why it becomes a sawtooth, etc etc. I also remember seeing regulators on the board but I do not see it in the schematic, where would they be present as per the schematic?
Max Current Power Supply – Burson Audio
Audio Myth - "Switching Power Supplies are Noisy" - Benchmark Media Systems
Marketing speak aside, the burson sounds very good and the whole amp is under 200$. I was curious how I could go about replicating this design. I understand that this is kind of an IC/PCB that might need more precision, but I'd love to know the engineering that goes behind all these, and if possible, diy options (atleast a design simulation).
What I see common is that they use really high switching frequencies and I'm not sure how they do it exactly. Burson has a diagram in its page (the rectifier diagram looks wrong lol), but I am unsure what that MCPS would have inside. Asking to a few other engineers they also told that high switching frequencies would have noise other than just switching noise, like electromagnetic radiation etc. and some really complicated compensations in grounding etc are needed in the PCB.
Would love guidance on what are all the things to look around and how to design such an smps? I do understand basics of ac-dc conversion, but only from a utility standpoint, not from a fidelity standpoint (at best I only know about ripples).
On a side note, I'd also love to know what is happening at each stage on the burson fun schematic. I can understand the ac to dc schematic but I'm unable to understand how the first noisy dc is converted into pulses, why it becomes a sawtooth, etc etc. I also remember seeing regulators on the board but I do not see it in the schematic, where would they be present as per the schematic?
Last edited:
Low noise switched supply is very tricky to design, common mode noise and diff. mode noise must be carefully controlled at the sametime. Cleverscope wrote about their super low noise isolated supply for instrumentation:
https://www.cleverscope.com/files/The Cleverscope CS448 Development Journey.pdf
For low power I would start with push-pull, but the transformer symmetry and switch symmetry problems mentioned in the article is probably out of scope for most diyers...
https://www.cleverscope.com/files/The Cleverscope CS448 Development Journey.pdf
For low power I would start with push-pull, but the transformer symmetry and switch symmetry problems mentioned in the article is probably out of scope for most diyers...
The subject of SMPS is strewn with pro and anti arguments based on possible usage and topologies vs expectations. Just as linear there are tradeoffs.
Noise - yes, switching creates noise but if the noise is managed in the design you will have little to no noise. This can be shifted by the switching rate, reduced by the topology used and the operation (ccm vs dcm for example). Lastly filtering on the power both to the control chips and the main power delivery along with the input.. means you'll end up with a quiet SMPS.
Switching speed
high speed = low inductance thus smaller inductor within reason, noise plus subharmonics (dcm) and harmonics are shifted out of the audio spectrum. Down side is the current over time and the inductor resistance to building the needed flux in that time period needs careful consideration. Noise easier to filter.
low speed = high inductance, thus larger inductor (again within reason for power requirement), noise in the audio spectrum making it harder, easier on the components and design meaning cheaper components.
To give an idea a 20A 15mH inductor weighs 32Kg. So typically you will not be running to the design calculations for the idea inductor.. perhaps 20% of that or less.
Inductor is the driver of the system so this needs to have enough stored flux energy per cycle = current * inductance / time period ..
Then there's duty rate - this causes stability problems above 50% requiring additional design.
Then you have the transients.. so your 200-300V becomes a 900-1.2kV component which then causes problems with capacitance. Although it's possible for a 900V mosfet to complete a switch on and off cycle in less than 100nS.
However... safe area of operation comes into effect (which is also related to switch speed).. you'll find that you may quickly cause overheating of the junction leading to demise of your switch.
Oh if the current in the inductor reaches zero on each pulse then that's DCM mode.. when the current hits zero and starts from zero it creates additional harmonics... but it's more efficient in operation..
In CCM the current never reaches zero in the inductor and oscillates with each pulse. A bit like A class vs B class as the current doesn't hit zero it creates very little noise but the current is causes heating and is wasted. So getting a CCM SMPS is low efficiency but quiet.. It's possible to run CCM in low load but then allow the current to max out when needed to give full power.
That's just boost topology..
Most of the time Boost is only used for lower wattage, flyback is used and can provide isolation at about the same wattage, but as soon as you get to 150-250W you may want to start looking at different topologies especially to reduce losses. 250-500W is half bridge and 500-1000W+ is full bridge, essentially driving power into a centre tapped transformer to cause an AC at the switching speed.. ie 100KHz which then means rectification by a half bridge or full bridge then occurs at 100Khz so the noise can be easily filtered out.
Only difference is the size of the transformers and the complexity of the control circuitry.
I would *strongly* recommend that any SMPS used for audio is *isolated* from the mains power. In the boost idea I'm attempting - I'm using an off the shelf SMPS (medical) that is isolated, filtered and has active power control to generate the LVdc. This means that the LVpsu has to pass all the transient safety tests. It also gives me a nice 150mVpp ripple starting point. Then the non-isolated boost is still isolated from the mains, but LV psu isn't isolated from the HVdc except for two fast switching 1.2Kv 20A diodes.
I could make a full PSU from 240V, through an active power factor correction into an isolating transformer, then through a regulator and onto a step up half bridge/flyback SMPS at a higher switch rate. Note this is precisely what your links are doing. They're probably using a lot of capacitors to minimise the ripple that needs to be supported by the PSU steps. The nice thing is the switching can also manage the power inrush for large capacitance as part of it's normal operation.
So they're switching up the frequency, filtering as they go, and then the final result is very low ripple and very little noise. If done right..
Noise - yes, switching creates noise but if the noise is managed in the design you will have little to no noise. This can be shifted by the switching rate, reduced by the topology used and the operation (ccm vs dcm for example). Lastly filtering on the power both to the control chips and the main power delivery along with the input.. means you'll end up with a quiet SMPS.
Switching speed
high speed = low inductance thus smaller inductor within reason, noise plus subharmonics (dcm) and harmonics are shifted out of the audio spectrum. Down side is the current over time and the inductor resistance to building the needed flux in that time period needs careful consideration. Noise easier to filter.
low speed = high inductance, thus larger inductor (again within reason for power requirement), noise in the audio spectrum making it harder, easier on the components and design meaning cheaper components.
To give an idea a 20A 15mH inductor weighs 32Kg. So typically you will not be running to the design calculations for the idea inductor.. perhaps 20% of that or less.
Inductor is the driver of the system so this needs to have enough stored flux energy per cycle = current * inductance / time period ..
Then there's duty rate - this causes stability problems above 50% requiring additional design.
Then you have the transients.. so your 200-300V becomes a 900-1.2kV component which then causes problems with capacitance. Although it's possible for a 900V mosfet to complete a switch on and off cycle in less than 100nS.
However... safe area of operation comes into effect (which is also related to switch speed).. you'll find that you may quickly cause overheating of the junction leading to demise of your switch.
Oh if the current in the inductor reaches zero on each pulse then that's DCM mode.. when the current hits zero and starts from zero it creates additional harmonics... but it's more efficient in operation..
In CCM the current never reaches zero in the inductor and oscillates with each pulse. A bit like A class vs B class as the current doesn't hit zero it creates very little noise but the current is causes heating and is wasted. So getting a CCM SMPS is low efficiency but quiet.. It's possible to run CCM in low load but then allow the current to max out when needed to give full power.
That's just boost topology..
Most of the time Boost is only used for lower wattage, flyback is used and can provide isolation at about the same wattage, but as soon as you get to 150-250W you may want to start looking at different topologies especially to reduce losses. 250-500W is half bridge and 500-1000W+ is full bridge, essentially driving power into a centre tapped transformer to cause an AC at the switching speed.. ie 100KHz which then means rectification by a half bridge or full bridge then occurs at 100Khz so the noise can be easily filtered out.
Only difference is the size of the transformers and the complexity of the control circuitry.
I would *strongly* recommend that any SMPS used for audio is *isolated* from the mains power. In the boost idea I'm attempting - I'm using an off the shelf SMPS (medical) that is isolated, filtered and has active power control to generate the LVdc. This means that the LVpsu has to pass all the transient safety tests. It also gives me a nice 150mVpp ripple starting point. Then the non-isolated boost is still isolated from the mains, but LV psu isn't isolated from the HVdc except for two fast switching 1.2Kv 20A diodes.
I could make a full PSU from 240V, through an active power factor correction into an isolating transformer, then through a regulator and onto a step up half bridge/flyback SMPS at a higher switch rate. Note this is precisely what your links are doing. They're probably using a lot of capacitors to minimise the ripple that needs to be supported by the PSU steps. The nice thing is the switching can also manage the power inrush for large capacitance as part of it's normal operation.
So they're switching up the frequency, filtering as they go, and then the final result is very low ripple and very little noise. If done right..
Last edited:
Thank you very much for this wonderful reply. Regarding CCM vs DCM, is the noise in DCM very similar to crossover noise we get from class B circuits (transistor non linearity with low level signals) or does it have a different profile (owing to inductor + capacitor pair)? I don't know which is the reason because we have both a switching element (gates) and a inductor+capacitor pair in these setups.
Do you think it will be necessary to learn about characterization of flyback transformers (or inductor+capacitor with good transient response) for this purpose. Would love to know if there are resources on the same. Regarding the bandwidth for drive circuit, can cascoding help or is there a catch (load compatibility etc)?
Also, what about rf noise, considering the high frequency of operations being carried out and the design considerations for pcb routing and ground planes. Is it similar to design of digital circuits?
I'm mainly into dacs and headphone amps so it will be well within 10W of peak power.
Do you think it will be necessary to learn about characterization of flyback transformers (or inductor+capacitor with good transient response) for this purpose. Would love to know if there are resources on the same. Regarding the bandwidth for drive circuit, can cascoding help or is there a catch (load compatibility etc)?
Also, what about rf noise, considering the high frequency of operations being carried out and the design considerations for pcb routing and ground planes. Is it similar to design of digital circuits?
I'm mainly into dacs and headphone amps so it will be well within 10W of peak power.
Last edited:
DCM is essentially the cycle has the inductor current to reach zero before starting the next charge. It can be done by static switch rate or if your controller has a DCM sense pin it may simply skip or adjust timing to achieve that.
The capacitor has nothing todo with DCM other than providing low impedance source of power (input caps with low ESR paralleled) and output which are typically to reduce ripple (and if a resistor is used - a low pass filter). If the output cap isn't there you'll get a switching sized ripple.
Oh I forgot - controlling current using switching has it's own issues with regard to switching. If you vary the switch rate (ie frequency) then the system needs to have filtering that will cope with the change of noise frequencies. No point designing and tuning filtering statically then only for the selected controller to change frequency based on the power needed.
The same occurs for situational triggering (ie light load) where the blanking or skipping of switching results in a different frequency noise.
There are multiple topologies - including synchronised switching (ie having a switch at the top and bottom), unsynchronised, also switching that removes the diode completely for the output (this reduces voltage drop) but you have to be careful of starting and shutdown scenarios where voltages and capacitors may cause negative polarity etc so design that in..
Personally I would have less trust in a mosfet than I would two 1.2kV in series. However there's also noise for each time a diode is used, same with a mosfet.
PCB design is probably better treated as a high power digital system. The PCB needs careful design and selection - having 10-20A through PCB traces means the right thickness, thermal design that doesn't cause heating of the tracks (ie 20A is something like 2.5mm minimum on the surface but 6.5mm under the surface to reduce heating caused by the current and the copper resistance.
If you're switching at 200KHz you will have 1MHz and up noise, you will have large fields created by the high current AC affecting everything near it and lastly you will have heat.
So you're likely to have different design and selection criteria depending on if you're making a high voltage/low current (valve amps) vs a low voltage/high current (solid state) vs a high voltage/high current (OTL low impedance) and that means one topology isn't suited for every scenario.
Last edited:
Thank you very much. I will try to learn further upon these scenarios and solutions. The amp that I have has a feedback loop that I think is controlling the switching rate. So it should be having filters that adaptively change to remove noise from it right? Any specific terms or design implementations I could search to get further detail on the implementation of such adaptive filters.
There's a lot of reading I need to do but so far the last but one paragraph relating to synchronised and unsynchronized switching is something I am able to relate a little bit to noise shaping and other design choices relating to delta Sigma modulators atleast with respect to the basics.
One more doubt, these would also need a stable clock right? Or is it fully self stabilized using these feedback loops?
There's a lot of reading I need to do but so far the last but one paragraph relating to synchronised and unsynchronized switching is something I am able to relate a little bit to noise shaping and other design choices relating to delta Sigma modulators atleast with respect to the basics.
One more doubt, these would also need a stable clock right? Or is it fully self stabilized using these feedback loops?
Last edited:
To answer your question on fly back etc - it's the iceberg problem. It's a bit like not being able to see the size, complexity of the problem when you're starting out and looking at a design. I wouldn't consider flyback until you've understood what are the basic problems they're solving, what each of the topologies does and the issues.
Choosing the wrong path will end with blown caps, mosfets at a minimum.. worse case death, destruction of property from fire etc.
So from a cost perspective - it may be more cost effective buying a commercial off the shelf unit.
It's possible (with the correct design) for 48V SMPS to be daisy chained in series so they produce higher voltage but same current limit or in parallel for more current with the same voltage. I know people here have daisy chained Cisco 48V SMPS together (because they're designed to support hot swap so in support parallel and in series connections). However not all SMPS can be connected like this - it depends on the topology.
Choosing the wrong path will end with blown caps, mosfets at a minimum.. worse case death, destruction of property from fire etc.
So from a cost perspective - it may be more cost effective buying a commercial off the shelf unit.
It's possible (with the correct design) for 48V SMPS to be daisy chained in series so they produce higher voltage but same current limit or in parallel for more current with the same voltage. I know people here have daisy chained Cisco 48V SMPS together (because they're designed to support hot swap so in support parallel and in series connections). However not all SMPS can be connected like this - it depends on the topology.
Is this relating to instability and growing oscillations, like a differential amplifier with feedback getting blown up if the the differential ends are shorted?
Thanks for the notice on flyback transformers. I'll keep it aside till I finish learning the basics. Have had some experience with basic buck/boost converters but only from a functionality standpoint for low frequency signals not much from a fidelity standpoint yet.
Thanks for the notice on flyback transformers. I'll keep it aside till I finish learning the basics. Have had some experience with basic buck/boost converters but only from a functionality standpoint for low frequency signals not much from a fidelity standpoint yet.
The amplification 'feedback' in the audio signal will have nothing todo with the SMPS.
Treat the SMPS as a completely separate component. You will then need to test that component with a range of scenarios - no load, low load, short circuit, normal load and high load, startup, shutdown, power transients from the mains power, etc etc.
So noise shaping is not just on the signal (ie the power pulses) but also into the support components. If not adequately filtered the noise can cause a steady DC supply line into the PWM chip to the reappear on the switch line, the switching frequency control and the current sensing. Ignoring the obvious over volt issues for a moment. Then that then extends to the PCB design.. which is another heap of problems.
I'm attempting to minimise the switching noise. Then shift that, then filter at the source, and finally.. a big RC 0.5Hz low pass filter made of 6,000uF 450V capacitance just to kill off any remaining noise or ripple.
Note that high frequency responses of capacitors and their impedance-frequency curve also needs review across the board (no easy just take a 50Hz cap... what's that cap do at 1Mhz, 10Mhz?
Thanks. I can now understand the complexity involved. I did try reading about real world responses of capacitors and inductors and their modelling (basic esr, esl is what I have seen till now), it sure was very complex and material dependant.
I think I would have a lot of learning to do, thanks for all the guidance. I'll try to experiment a little and also get through some books on these topics. Hope to get back with a design that I can share for review and get my mistakes corrected.
I think I would have a lot of learning to do, thanks for all the guidance. I'll try to experiment a little and also get through some books on these topics. Hope to get back with a design that I can share for review and get my mistakes corrected.
To be honest I think you are biting off way more than you can chew in a reasonable amount of time.
There is a reason why even test equipment manufacturers with dozens of highly qualified electrical engineers often decide to outsource SMPS development to specialised manufacturers or make use of consulting/design services. This isn't a matter of reading a few books or App notes, it's more of a career path on its own. Not saying you won't be able to come up with a reasonable design at some point, but it will probably be more expensive and worse performance than something you can buy off the shelf right now, QC tested, with warranty.
There is a reason why even test equipment manufacturers with dozens of highly qualified electrical engineers often decide to outsource SMPS development to specialised manufacturers or make use of consulting/design services. This isn't a matter of reading a few books or App notes, it's more of a career path on its own. Not saying you won't be able to come up with a reasonable design at some point, but it will probably be more expensive and worse performance than something you can buy off the shelf right now, QC tested, with warranty.
I think there is even a few forums focussed only on power conversion topics. I don't know about the quality of content there, but they exist. Maybe worth to check them out.
Edit: TI also has some reference designs, but you will probably have to wind your own transformers if you would want to replicate any of them because I doubt the transformers with the adequate turns ratios are available off the shelf in low quantities.
PMP40379 500-W, single stage LLC power supply reference design for audio amplifier | TI.com
PMP20195 Off Line Power Supply for Audio Reference Design | TI.com
Edit: TI also has some reference designs, but you will probably have to wind your own transformers if you would want to replicate any of them because I doubt the transformers with the adequate turns ratios are available off the shelf in low quantities.
PMP40379 500-W, single stage LLC power supply reference design for audio amplifier | TI.com
PMP20195 Off Line Power Supply for Audio Reference Design | TI.com
Last edited:
Thanks. I'll surely check them out. When I started this thread I was mainly curious about what made the amplifier I bought sound so good and thought maybe I could diy a similar setup. I didn't understand the complexity of it, since the marketing page just had black boxes and a feedback loop and just mentioned all the magic is in their power supply.
Currently I'm trying to diy a dac design, which seems far more doable with my current knowledge (still not trivial, but atleast I think I can do it). Pcb design is one area I'm not quite comfortable with yet due to lack of prior experience but I got guidance on getting started and hopefully I would be able to make decent result in due course of time.(https://www.diyaudio.com/forums/dig...c-pcb-design-guidelines-help.html#post6585209)
Edit: I have read a little about transformer design but winding my own one would be another level of complexity for the moment and very hard for me to comprehend (Magnetism). I do know a company that engages in such complex designs maybe I could seek guidance from them when I reach that stage.
Currently I'm trying to diy a dac design, which seems far more doable with my current knowledge (still not trivial, but atleast I think I can do it). Pcb design is one area I'm not quite comfortable with yet due to lack of prior experience but I got guidance on getting started and hopefully I would be able to make decent result in due course of time.(https://www.diyaudio.com/forums/dig...c-pcb-design-guidelines-help.html#post6585209)
Edit: I have read a little about transformer design but winding my own one would be another level of complexity for the moment and very hard for me to comprehend (Magnetism). I do know a company that engages in such complex designs maybe I could seek guidance from them when I reach that stage.
Last edited:
If you want to learn about PCB layout topics, maybe Robert Feranec's YT channel is of interest to you. He has interesting talks/interviews with Rick Hartley, Eric Bogatin and others.
Most of the content is more targeted at high-speed digital design, but mixed-signal design was also touched upon recently. You also received some good pointers in your other thread, especially the one from Pavel (BesPav).
Most of the content is more targeted at high-speed digital design, but mixed-signal design was also touched upon recently. You also received some good pointers in your other thread, especially the one from Pavel (BesPav).