It is reasonable to assume that the peak current demand is in the range 2times to 5times the nominal current delivered to a resistive load driven to full power.
i.e. a 100W into 8ohms amplifier will see a peak demand in the range 10Apk to 25Apk compared to 5Apk into 8r0 dummy load.
I design for 3times, i.e. 15Apk in a 100W into 8ohms amplifier.
An SMPS of ~ 200W @ a regulated +-45Vdc will probably NOT be able to meet that 15Apk speaker demand.
A 1350W SMPS is guaranteed to meet a 15Apk demand when regulated to +-45Vdc supply voltage.
An SMPS somewhere between 200W and 1350W will meet the normal peak transient demands of some severe reactance speakers.
What end of that range do we need to aim for?
Last edited:
Nigel, where did you get the coeffs for your 8th order Bessel & Butterworth filters .. and how did you translate them into resistors & caps?
_________________
Just a reminder to everyone ..
ALL reliable Double Blind Listening Tests bla bla on Bandwidth Limitation (including some of mine) show clear preference for Bandwidth Limitiation to 20kHz & below by those who can tell the difference.
The tests done this Millenium confirm those in the previous century.
Of course loadsa Golden Pinnae think differently but the DBLT will soon sort out these deaf pseudo golden pinnae cos they won't be able to reliably pick out the difference
_________________
So Nigel's filter is useful, not only to those measuring Class D amps .. but also to those who want their conventional systems to sound better.
I will explain my thoughts . One writer said and was born out by experiment that JFET's were better ( much ) . That set off a chain of thoughts . Starting with passive RIAA circuits . I was told years ago by a friend that the superiority of the inverting input is that the gain formula could arrive at 0 and mostly match a passive circuit . Now to the crunch . Ticks on records are said to extend in active RIAA circuits . My friend argued that it is not automatically true if inverting ( these were the days of 741 ) . I reasoned , that being so it should slightly outperform a passive version and was the better gamble . I suspect in truth it will rest on a knife edge which is better . Looking at the work involved it seemed better to use 2 x 4 op amps . The layout of quad op amps is wonderful for this . I was convinced the multiple feedback pairs I gave in the beginning would ring .
I found the gain problem mentioned in the first post . I had a 3k9 and 39 K input resistor . They were bridged . The real audio band gain is 510 to 512 mV , nearly perfect . I have the Hypex running from +/- 37 VDC . It gives 23.5 Vrms at clipping ( using 39 K input giving 2.35 V out ) . That is about 3.8 V lost ( 66.46 V pk to pk out of 74 ) .
Using the FET op amp in and bipolar out proved to be a good choice . Very cheap also . GBP TLO 74 3 MHz MC33079 16 MHz . 33078 is like a nicer 5532 with less output current , 33079 is the quad package . 33079 looks to have low distortion if the graphs are correct .
I have proof that it is a good general device as the schools type test oscillator I use it about 1% THD . It possibly uses a DAC . The clean up works on it also . Look at the Hypex graphs below . The output is less distorted than the input . I can only guess this is as in valve amp stages cancellation of curve shape . No worries as for what I need to do it works well enough . - 3db point is 47 kHz of the filter . No unexpected bumps . I suspect the standard option filter using 4 op amps would have added it's own problems . Certainly the bipolar input did ( ringing ) . The high distortion oscillator is great in having virtually no voltage difference at any frequency . I use a Wien bridge oscillator ( NE5532 ) plus filter circuit with PA 53 thermistor for the more detailed tests . It needs some work as I am no longer looking at valve amps now .
The choice of 186 kHz was that it was as high as I could go with any chance of doing anything good . If reset to 100 kHz doubtless it would help in some areas . Some DIL switches perhaps .
The distorted graph is Hypex with no filter . The filter graph removes more than the Hypex distortion . The schools type oscillator obviously has similar problems . The final result is ultra clean . The indicated problems can not be in-band as the filter only stats to work at 47 kHz . The linearity of the filter is great so I have no doubts .
Might I suggest what we see might be what we hear ?
That is correct, but how is that a limitation? There is no need to advocate the advantages of a raw power supply over a regulated power supply, clearly the raw power supply has advantages for a class AB audio power amplifier (not to mention the price/performance factor).
I thought the discussion is specifically about SMPS and the need of "tens of thousands uF" at the output. I still don't think this is required, if the SMPS control loop has enough bandwidth, as much as a wide bandwidth linear regulated power supply also doesn't need a gazillion of uF at the output.
I'm also saying that the load current transient behaviour (that is, the output voltage drop/spike for a fast load current variation) depends mostly on the control loop bandwidth rather than on the amount of output uF's. This applies for both SMPSs an linear PSs.
I agree with Bob. A conventional iron cored transformer PSU tends to have a better short-term overload margin than a SMPS of the same nominal power rating.
Hanging large capacitors onto the SMPS output is one solution. Another is just to design the SMPS to allow short-term overloads. The magnetics in a SMPS have some thermal capacity (albeit less than a big iron-cored transformer) so they can be overloaded for a short time without too much trouble. The main issue has historically been the switching MOSFETs which are easy to burn out with excessive current, but cheap ultrafast IGBTs are now available that handle overloads much better.
I expect PA amp designers have been doing something like this for years already.
Hanging large capacitors onto the SMPS output is one solution. Another is just to design the SMPS to allow short-term overloads. The magnetics in a SMPS have some thermal capacity (albeit less than a big iron-cored transformer) so they can be overloaded for a short time without too much trouble. The main issue has historically been the switching MOSFETs which are easy to burn out with excessive current, but cheap ultrafast IGBTs are now available that handle overloads much better.
I expect PA amp designers have been doing something like this for years already.
And your point is? If you don't like the results, then don't use a regulated power supply (SM or linear) and stick with a raw power supply (which is anyway recommended, for precisely that reason).
A large output capacitor won't help, anyway, unless the regulator control loop HF performance is so poor that it's output impedance (small and large signal) is much higher than the impedance (including the ESL and ESR) of the gazillion uF output capacitor. Only then the output capacitor(s) will deliver the required current, but that would prove nothing but a poor designed power supply.
This is not about any control loop issue, it is about the instantaneous current that can be supplied. If you use a 200W SMPS for a +/-40V 100W amp, you obviously cannot supply the required peak currents with some reactive speakers which may be nominally 5A but, as Andrew argued (and I agree) can easily be 10 or 20A short-term. For your nominal 200W SMPS to be able to supply that, it would need some reservoir cap at its output to absord that transient load. Whether it would need to be a zillion is debatable.
Without any reservoir cap, an SMPS that can supply up to 20A at +/-40V would need to be 800W or more. A couple of caps with a 200W supply seems more sensible to me.
Jan
Here are a few things to consider regarding reservoir capacitors for SMPS used in audio power amplifiers.
In my earlier post I was referring to the reservoir capacitor at the output side of the SMPS, across the amplifier rails. This is where you may often find a 50,000uF cap on a good audiophile amplifier. Lesser quality amplifiers with conventional power supplies may sometimes use as little as 10,000uF, but my focus here is on audiophile designs.
The intermediate bus reservoir capacitor on the mains side of the SMPS must also be considered. This is where the mains voltage is typically full-wave rectified to feed the heart of the isolated SMPS switcher. This capacitor must also be fairly large in a good power amplifier. The intermediate bus voltage will be on the order of 170V or more in a typical SMPS designed for 120V mains (it will be higher in an SMPS with a power factor correction circuit). The intermediate bus supply operates at the mains frequency, so it needs substantial capacitance to deal with ripple. Consider a modest amplifier that consumes 500W from the mains at full power. This corresponds to about 3A at 170V. Ripple on the intermediate bus should be held to less than 1V. This requires a 24,000uF intermediate bus reservoir capacitor rated at probably 250V or more. Not a cheap capacitor.
In an amplifier with PFC the intermediate bus capacitor can often be smaller, since the PFC boost converter switches at a high frequency.
This might be all you would need if the intermediate bus was followed by an ideal isolated SMPS to create the rail voltages. But that is not the real world.
Large SMPS typically run at no more than 1MHz, usually 500kHz, even nowadays. It is true that low-voltage point-of-load Buck converters can operate at much higher frequencies, but that is not generally the case for power amplifier SMPS. A rule of thumb for the SMPS step current recovery time constant is ten times the switching period. So the control bandwidth may be on the order of 50kHz. A good reference here is “Switching Power Supply Design” by Pressman, Billings and Morey, McGraw-Hill, 2009.
A control bandwidth of only about 50kHz is simply insufficient to maintain adequate broadband low impedance of the SMPS for a high-quality power amplifier. Significant reservoir capacitance is thus needed, even in this small-signal sense. However, a 1000uF reservoir capacitor will get the rail impedance below 10 miili-ohm at 20kHz, which may be adequate in the small-signal sense.
In the large-signal context, the power supply must be able to supply the peak load current demanded of the loudspeaker via the rails. This can in theory be quite high. Ideally, an amplifier should be able to deliver the maximum current demanded by a load that consists of the DC resistance of the load in series with an arbitrary reactive network when it is driving the load with an arbitrary waveform band-limited to perhaps 50kHz. In some loudspeakers that DC resistance might be as low as 2 ohms. A simple example of the high currents possible can be seen in Figure 22.6 in my book. There a nominal 8-ohm speaker load with a 6.4-ohm DC resistance is driven by a 28V-peak rectangular waveform constructed to maximize peak current. Peak current reaches 10A, while the corresponding peak current for this 50-watt amplifier into an 8-ohm load would be about 3.5A.
Unless it is way over-built, an SMPS cannot handle such large peak currents without a reservoir capacitor. Note that when an SMPS max’s out, it goes out of regulation and its control loop no longer enforces low impedance at the output.
Class AB amplifiers put a very strong and potentially asymmetrical current drain on the supply, with very high-frequency crossover components and a wide range of load current. The current waveform is like that of a half-wave-rectified version of the output signal current. The current might dynamically fall to an idle current of 150mA at crossover into a resistive load to 5A into an 8-ohm load at 40V at 100 watts. At 4 ohms, this will be 10A at 200 watts. It is easy to see that the dynamic current swing can be 60:1. Switching power supplies do not always work so well over such a wide range of load current. The large variation in load current at the audio rate will make the SMPS control loop work quite hard.
SMPS is the way to go for the very best quality audiophile amplifiers as long as they are designed as a system to deliver the highest peak current that will be required and not create RFI that will get into the amplifier electronics. Big advantages include greater isolation from mains noise and relatively constant rail voltages independent of mains voltage variations. Those incorporating PFC are even more immune to imperfections in the mains supply and are often able to effectively extract more power from the mains because they extract varying amounts of power from the mains over nearly the full cycle (rather than impulsive high-current bursts at cycle peaks).
Cheers,
Bob
In my earlier post I was referring to the reservoir capacitor at the output side of the SMPS, across the amplifier rails. This is where you may often find a 50,000uF cap on a good audiophile amplifier. Lesser quality amplifiers with conventional power supplies may sometimes use as little as 10,000uF, but my focus here is on audiophile designs.
The intermediate bus reservoir capacitor on the mains side of the SMPS must also be considered. This is where the mains voltage is typically full-wave rectified to feed the heart of the isolated SMPS switcher. This capacitor must also be fairly large in a good power amplifier. The intermediate bus voltage will be on the order of 170V or more in a typical SMPS designed for 120V mains (it will be higher in an SMPS with a power factor correction circuit). The intermediate bus supply operates at the mains frequency, so it needs substantial capacitance to deal with ripple. Consider a modest amplifier that consumes 500W from the mains at full power. This corresponds to about 3A at 170V. Ripple on the intermediate bus should be held to less than 1V. This requires a 24,000uF intermediate bus reservoir capacitor rated at probably 250V or more. Not a cheap capacitor.
In an amplifier with PFC the intermediate bus capacitor can often be smaller, since the PFC boost converter switches at a high frequency.
This might be all you would need if the intermediate bus was followed by an ideal isolated SMPS to create the rail voltages. But that is not the real world.
Large SMPS typically run at no more than 1MHz, usually 500kHz, even nowadays. It is true that low-voltage point-of-load Buck converters can operate at much higher frequencies, but that is not generally the case for power amplifier SMPS. A rule of thumb for the SMPS step current recovery time constant is ten times the switching period. So the control bandwidth may be on the order of 50kHz. A good reference here is “Switching Power Supply Design” by Pressman, Billings and Morey, McGraw-Hill, 2009.
A control bandwidth of only about 50kHz is simply insufficient to maintain adequate broadband low impedance of the SMPS for a high-quality power amplifier. Significant reservoir capacitance is thus needed, even in this small-signal sense. However, a 1000uF reservoir capacitor will get the rail impedance below 10 miili-ohm at 20kHz, which may be adequate in the small-signal sense.
In the large-signal context, the power supply must be able to supply the peak load current demanded of the loudspeaker via the rails. This can in theory be quite high. Ideally, an amplifier should be able to deliver the maximum current demanded by a load that consists of the DC resistance of the load in series with an arbitrary reactive network when it is driving the load with an arbitrary waveform band-limited to perhaps 50kHz. In some loudspeakers that DC resistance might be as low as 2 ohms. A simple example of the high currents possible can be seen in Figure 22.6 in my book. There a nominal 8-ohm speaker load with a 6.4-ohm DC resistance is driven by a 28V-peak rectangular waveform constructed to maximize peak current. Peak current reaches 10A, while the corresponding peak current for this 50-watt amplifier into an 8-ohm load would be about 3.5A.
Unless it is way over-built, an SMPS cannot handle such large peak currents without a reservoir capacitor. Note that when an SMPS max’s out, it goes out of regulation and its control loop no longer enforces low impedance at the output.
Class AB amplifiers put a very strong and potentially asymmetrical current drain on the supply, with very high-frequency crossover components and a wide range of load current. The current waveform is like that of a half-wave-rectified version of the output signal current. The current might dynamically fall to an idle current of 150mA at crossover into a resistive load to 5A into an 8-ohm load at 40V at 100 watts. At 4 ohms, this will be 10A at 200 watts. It is easy to see that the dynamic current swing can be 60:1. Switching power supplies do not always work so well over such a wide range of load current. The large variation in load current at the audio rate will make the SMPS control loop work quite hard.
SMPS is the way to go for the very best quality audiophile amplifiers as long as they are designed as a system to deliver the highest peak current that will be required and not create RFI that will get into the amplifier electronics. Big advantages include greater isolation from mains noise and relatively constant rail voltages independent of mains voltage variations. Those incorporating PFC are even more immune to imperfections in the mains supply and are often able to effectively extract more power from the mains because they extract varying amounts of power from the mains over nearly the full cycle (rather than impulsive high-current bursts at cycle peaks).
Cheers,
Bob
No output cap is not possible, at least because the control loop needs compensation.
Otherwise, I disagree, in a voltage regulator (linear or SMPS) the output caps do not need to deliver the short term current, if the HF output impedance of the regulator is low enough (which maps to a wideband control loop.
I am happy that you recognize the critical role of the control loop bandwidth in delivering large current transients.
Though, I would not consider 50KHz as a state of the art SMPS control loop bandwidth. Current SMPSs (in particular those with a PFC in front) can (and actually do) easily switch in the MHz range. Also, 1/10 of the switching frequency control bandwidth is IMO unreasonable low today, I've designed for 1/3 to 1/5.
If 10,000 uF or more of output capacitors are really needed, one may legitimately wonder what's the point of using a SMPS and not stick with raw power supplies. Supply voltage ripple and mains noise are easily rejected by a decent PSRR (not to mention that a new, internal, EMI source has to be dealt with).
Designing a 8ohm nominal amp to handle 2ohm load impedance dips is also a stretch. I've seen a number of commercial amps specified in 8, 4, 2, 1 ohm loads, and those amplifiers have exactly that huge (>2Kw) type of power supply you mentioned.
Waly, I get your point, but many an SMPS has a hard current limit of say 5A or 3A. That's part of the circuit protection. For such a case, you need reservoir caps to provide transient current.
I do not dispute the control loop importance but I think we talk about two different things.
Jan
Once the max current output is reached the SMPS will see its
output impedance rise sharply , almost a straight vertical line ,
the control loop reaction time only role will be to set the time it takes
from the surge to current limitation/increasing output impedance
to occur , i guess that it s what Jan is explaining.
I'm sorry, but a large reservoir cap won't avoid the current protection kicking in, at least in a good quality SMPS!
Again, the transient high current will come from the regulator and not the output cap, if the control loop is fast enough to keep the HF impedance low. That's in no way different from a linear regulator, you should know this from the Jung-Didden regulator. Therefore, the current limit will still apply, even if the output cap is large. The SMPS current limiting is also functionally no different from a linear regulator current limiting.
It is true that I have never seen a SMPS without current limiting circuitry (usually cycle-by-cycle based).
That is true, once the protection kicks in, the output switches from constant voltage to constant current, and the output impedance goes high.
Now, relying on a large output cap to avoid the voltage drop when switching to CC mode defeats the very purpose of the current limiting. Do we really want to discharge those "tens of thousands uF" in the speaker when a catastrophic failure occurs? After all, wouldn't be much simpler and cheaper to use a raw power supply and forget about regulated power supplies (SMPS or linear)?
Another good reason to use switch-mode is to close the gap between class D and class H ( G ) . For example 75 % + efficiency H and less good D about 86 % .
The Hypex SMPS 400 that I have would be ideal . One thing I want to try is a D Self Complimentary feedback pair class B using this PSU . TDA 2030 states 67 % efficient . I have a hunch I should be able to beat that . I usually accept 50 % .
My little test rig is working fine for Hypex ( 8 x 186 kHz filter with 2 x TL074 , last page before # 2100 ) . I even have some excellent 0.6 mW tests . I went back to all TL074 in the end for testing as it works best . Although MC33079 is said to be unity gain stable it is not so with 400 kHz presented . As someone said this FFT pre filter is very good with things other than class D . I will take the idea of using JFET inputs more seriously now . Maybe like valves they are better when RF is about ?
The Hypex SMPS 400 that I have would be ideal . One thing I want to try is a D Self Complimentary feedback pair class B using this PSU . TDA 2030 states 67 % efficient . I have a hunch I should be able to beat that . I usually accept 50 % .
My little test rig is working fine for Hypex ( 8 x 186 kHz filter with 2 x TL074 , last page before # 2100 ) . I even have some excellent 0.6 mW tests . I went back to all TL074 in the end for testing as it works best . Although MC33079 is said to be unity gain stable it is not so with 400 kHz presented . As someone said this FFT pre filter is very good with things other than class D . I will take the idea of using JFET inputs more seriously now . Maybe like valves they are better when RF is about ?
Agree with all this although you could put a low value resistance
between the SMPS and the caps , this will integrate the peak current
seen from the SMPS but then the regulation would be somewhat less
efficient not counting the likely increased output impedance at low
frequencies.
The raw PSU would still discharge big caps into the speaker
but it s true that circuits where currents can be held within
control are more desirable , in the case of SMPS a capacitor ,
or rather two // , totaling no more than 10 000 uF feeded
with said low value resistance seems to me a not so bad
compromise for who wants to go SMPS.....
Last edited:
"Practical high order Bessel & Butterworth filters for Dummies"?
Was it "Practical high order Bessel & Butterworth filters for Dummies"? It must be quite an advanced text as it includes a spec for the OPAs too.
Could you scan the relevant pages & post for the benefit of us unwashed masses? If it was a link .. even better.
Nigel, what I asked was how did you calculate your resistor & cap values for your filters.
Was it "Practical high order Bessel & Butterworth filters for Dummies"? It must be quite an advanced text as it includes a spec for the OPAs too.
Could you scan the relevant pages & post for the benefit of us unwashed masses? If it was a link .. even better.
Of possible value, a very nice, tight summary of considerations regarding attenuation of ripple and noise in SMPS ... www.ti.com/lit/an/snoa561a/snoa561a.pdf ...
SMPS for Golden Pinnae power amps
A spec for SMPS for Golden Pinnae power amps is emerging. It will have
Anyone know if the Hypex 'semi-regulated' SMPSs conform to the above.
___________________
BTW, its Class A amps that stand to gain the most audible improvement from the above. Also easier to design SMPS for Class A Golden Pinnae Class A amps.
A spec for SMPS for Golden Pinnae power amps is emerging. It will have
- Power Factor Correction : Less sh*t on the mains is nearly as good as having your mains cables supported by pine cones grown in Unicorn dung by Virgins.
- 'Semi-regulated' behaviour going from Constant Voltage to Constant Power when Current Demands exceeds a certain level. ie it doesn't Current Limit.
- Short term Peak Power delivery several times greater than the continuous rating
- Output capacitors are part of the above behaviour and integrated into the design of the SMPS. Jan.didden would have zillion uF. Bob Cordell probably only 1/2 a zillion as befits his objective engineering background. Waly would use just 1uF but his advanced knowledge of 21st century SMPS design compensates. But that 1uF is hand carved by Virgins from Solid Unobtainium
- There's loadsa design choices to do with how much the supply is allowed to sag on 'overload' but the important principle is that any sag is 'clean' and dun introduce nastiness on clean or overloaded signals. Wahab, no need for series resistors. You design this behaviour into your SMPS.
Anyone know if the Hypex 'semi-regulated' SMPSs conform to the above.
___________________
BTW, its Class A amps that stand to gain the most audible improvement from the above. Also easier to design SMPS for Class A Golden Pinnae Class A amps.
Last edited:
The purpose of the SMPS current limiting is to save the SMPS, has nothing to do with catastrophic failure.
If the SMPS output collapses, which is does once the max current is reached, you need those reservoir caps to keep the amp going.
Its functionally not different from a rectifier/capacitor supply. The diodes disconnect the reservoir caps from the transformer about 80% of the time. During those 80%, you rely on the reservoir caps to keep the amp going. Same thing.
Jan
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.