Today I spent some time and did some simulations on the power amps I have built and in the process of rebuilding. The ISP and VAS stages use current sources and cascodes as usual, and the OPS stage uses L-MOSFETs source follower.
The following are the result:
IPS + VAS PSRR
20Hz -> -69dB
50Hz -> -68dB
100Hz -> -67dB
200Hz -> -66dB
500Hz -> -63dB
1,000Hz -> -59dB
2,000Hz -> -56dB
5,000Hz -> -45dB
10,000Hz -> -39dB
20,000Hz -> -33dB
50,000Hz -> -25dB
100,000Hz -> -19dB
OPS PSRR
200Hz -> (unable to measure)
500Hz -> -101dB
2,000Hz -> -78dB
20,000Hz -> -57dB
I am not sure how accurate the models are as I got them from a friend. The figures are a bit rough but are indicative.
With the above simulation results, I have come up with a passive LCR filter to be installed for the front end (i.e. IPS and VAS) rails, which will achieve -44dB at 20Hz and much better higher up. This will make the front end to have better than -110dB PSRR in the entire audio band.
For the OPS rail, in order to achieve -110dB PSRR I need the power supply to provide -53dB. In other words, the power supply impedance at 20kHz must be 0.00224R or lower. This will be very difficult to achieve. Using 14 x Rubycon ZLJ 680uF/100V close to the output L-MOSFETs on a power and ground plane may help achieve this figure. But I think that the two layer PCB is very difficult to design. I don't think I can make that number. If I can manage to have the impedance seen by the L-MOSFETs to be 0.01R then I would be achieving only -93dB at 20kHz.
But before we go any further, we must first establish what total PSRR (20Hz to 20kHz) of an amplifier is necessary for good sound. I have had no ideas at all and came up with only an arbitary number of -110dB. -140dB was previously suggested by others.
I hope you guys can continue the discussions and come up with something. I will be away overseas from tomorrow and will be back in May.
Regards,
Bill
The following are the result:
IPS + VAS PSRR
20Hz -> -69dB
50Hz -> -68dB
100Hz -> -67dB
200Hz -> -66dB
500Hz -> -63dB
1,000Hz -> -59dB
2,000Hz -> -56dB
5,000Hz -> -45dB
10,000Hz -> -39dB
20,000Hz -> -33dB
50,000Hz -> -25dB
100,000Hz -> -19dB
OPS PSRR
200Hz -> (unable to measure)
500Hz -> -101dB
2,000Hz -> -78dB
20,000Hz -> -57dB
I am not sure how accurate the models are as I got them from a friend. The figures are a bit rough but are indicative.
With the above simulation results, I have come up with a passive LCR filter to be installed for the front end (i.e. IPS and VAS) rails, which will achieve -44dB at 20Hz and much better higher up. This will make the front end to have better than -110dB PSRR in the entire audio band.
For the OPS rail, in order to achieve -110dB PSRR I need the power supply to provide -53dB. In other words, the power supply impedance at 20kHz must be 0.00224R or lower. This will be very difficult to achieve. Using 14 x Rubycon ZLJ 680uF/100V close to the output L-MOSFETs on a power and ground plane may help achieve this figure. But I think that the two layer PCB is very difficult to design. I don't think I can make that number. If I can manage to have the impedance seen by the L-MOSFETs to be 0.01R then I would be achieving only -93dB at 20kHz.
But before we go any further, we must first establish what total PSRR (20Hz to 20kHz) of an amplifier is necessary for good sound. I have had no ideas at all and came up with only an arbitary number of -110dB. -140dB was previously suggested by others.
I hope you guys can continue the discussions and come up with something. I will be away overseas from tomorrow and will be back in May.
Regards,
Bill
Last edited:
This post'll be buried by May, but the main piece you're missing is the control loop will attenuate output stage errors by its excess loop gain. Provided control loop is implemented with sufficient PSRR that it's own "noise" floor with respect to ripple on the rails is low enough for it to see those errors. In a traditional discrete power amp with Miller compensation and ~25dB gain achieving 50+dB excess loop gain at 20kHz is not trivial. This is part of the reason why people find audible differences in current feedback amps and feedforward compensation. A simpler solution's to use an IC with good GBP; 50MHz parts like the LME49710 and LME49811 have about 70dB loop gain at 20kHz.
The other piece is the analysis I linked some pages back in this thread, which yields required PSRR as a function of bypass capacitance. For practical reservoir sizes, typical amp quiescent currents, and typical signal levels the supply ripple will be around 30dB larger than the signal. So upwards of 90dB PSRR in the control loop is wanted through some combination of filtering, regulation (usually more attractive than filters), and the control devices' own PSRR. 110dB is not a bad default as 90dB is arguably rather minimal. 140dB is overkill but trivial to achieve on paper when using IC regulators and control devices. In practice, laying out a PCB to hit an O(100nV) noise floor is not so easy and quiet control devices in a loop with low noise gain are needed (LME49990 or AD797 at unity gain, for example).
The other piece is the analysis I linked some pages back in this thread, which yields required PSRR as a function of bypass capacitance. For practical reservoir sizes, typical amp quiescent currents, and typical signal levels the supply ripple will be around 30dB larger than the signal. So upwards of 90dB PSRR in the control loop is wanted through some combination of filtering, regulation (usually more attractive than filters), and the control devices' own PSRR. 110dB is not a bad default as 90dB is arguably rather minimal. 140dB is overkill but trivial to achieve on paper when using IC regulators and control devices. In practice, laying out a PCB to hit an O(100nV) noise floor is not so easy and quiet control devices in a loop with low noise gain are needed (LME49990 or AD797 at unity gain, for example).
Last edited:
Twest820, Thanks for your post. I will fly tonight but before that I will come back here when I can.
My amp uses two pole Miller compensation. The amp has a close loop gain of just over 30dB. OLG is 87dB at 100Hz and falls down to 70dB at 20kHz. So it has 40dB excess loop gain at 20kHz. I don't know if that is good enough. I don't have the knowledge and skills for amplifier design and I only play with other stuff like power supply, etc. Is the amp good enough? or poor?
I derived the above listed figures in close loop so I think they should include everything, such as feedback correction, etc. I created two perfect DC sources, one for the front end and the other for the backend, and created a voltage source of 8V peak to peak in varying frequencies and put it directly on the rail, separated by small R and H from the original perfect DC sources. I then injected the maximum allowable signal before clipping (1.8V) at 10Hz at the input of the amplifier and then checked the ripples on the output of the amplifier. The noise at the output then compared to the rail noise of 8V.
My amp uses two pole Miller compensation. The amp has a close loop gain of just over 30dB. OLG is 87dB at 100Hz and falls down to 70dB at 20kHz. So it has 40dB excess loop gain at 20kHz. I don't know if that is good enough. I don't have the knowledge and skills for amplifier design and I only play with other stuff like power supply, etc. Is the amp good enough? or poor?
I derived the above listed figures in close loop so I think they should include everything, such as feedback correction, etc. I created two perfect DC sources, one for the front end and the other for the backend, and created a voltage source of 8V peak to peak in varying frequencies and put it directly on the rail, separated by small R and H from the original perfect DC sources. I then injected the maximum allowable signal before clipping (1.8V) at 10Hz at the input of the amplifier and then checked the ripples on the output of the amplifier. The noise at the output then compared to the rail noise of 8V.
Last edited:
twest820, how right you are!!!!!!!
I have just tested my passive filter for the front end. While it does a great job of eliminating ripples from the rails, it does not work well because of the loads and the filter's impedance generating its own ripples.
So now I will need to look hard to find a 3 terminal regulator that does +/-90VDC.
Yes - I was inspired by that work of Terry's when he posted it - I could not though duplicate his lack of series inductance with my setup. I was using an LCR meter to measure though and he used a network analyser. Frank (fas42) has since pointed me to a couple of Terry's posts on my blog. I find myself in agreement with Terry - why don't amplifier manufacturers adopt this approach? For myself I'm keen to explore 3D capacitor wiring solutions before I settle on a 2D one like his.
Yes - I have some (quite a few!) very close-in SMT ceramics soldered on to the pins of the chipamp, so inductance isn't too much of a worry. If inductance was a problem I reckon the FFT of the PSU noise would show bumps at higher freqs - but in fact its decreasing monotonically as freq rises. With the X5Rs in circuit the challenges are getting the mid-band and LF PSU noise down.
Yep its a great scheme and I built one. I may end up building more, we shall see...
@HiFiN - not to rain on your parade but why terminate your PSRR analysis at 20kHz? Since the rail currents are haversines for a classAB amp, there are going to be plenty of harmonics above 20kHz on the supply. I reckon those get inside the control loop when PSRR is inadequate and will introduce IMD, greying out the perceived sound
Last edited:
I look at it until about 100MHz when looking at power supply line filtering, and that is why I am always interested in using some RF chokes as they may be more effective than capacitors. But those are for filtering the line, rail / switching noises or EMI pick up. I guess that amplifier signals would not get higher than 40kHz so the load generated noises on the rails would not be much higher in frequencies. Here I guess that low impedance is what is needed. In terms of low impedance, modern electrolytic capacitors have impedance quite flat from about 20kHz to 200kHz. So if the impedance is low enough for 20kHz, it is low enough fro 200kHz. Above that frequency, perhaps low impedance is less important because there is no signal generated noise, but filtering still is. Filtering high frequencies is easy. Providing a low impedance at high frequency is not. I think they are two separate things. So in a word, for filtering noises, I look into the MHz region. For providing a low impedance to the load, perhaps 20kHz is sufficient because in that case up to 200kHz is automatic and even 1MHz is still pretty good.
Of course, I may be wrong here because perhaps the low impedance is needed until the unity gain point of the amplifier?
Last edited:
The folks on the solid state forum could probably give a better answer than I could even if I had the schematic and loop gain plot, but it's reasonable. Your own math says you want 55dB, so presumably it's not good enough.
Well, one way to solve that would be to ask if you really need a 500W/channel amp and, if not, drop the rails to what you do need. If that gets you under 40V at the high mains corner, well, problem solved. If not, here would be a good starting point for Maida options. I'm not aware of any negative builds but the main difficulty in flipping the floating topology to the negative rail is finding a suitable PNP pass device.So now I will need to look hard to find a 3 terminal regulator that does +/-90VDC.
For the positive rail there's also the TL783. And the dual positive topology if you've a winding available.
The DAC probably doesn't stop at 20kHz either---if it's a slow rolloff type it'll go to 35kHz or so with redbook---but generally if something's well designed at 20kHz it holds up OK quite a ways higher. This is one the reasons a 90dB PSRR is a minimal target; if you design to, say, 110dB at 20kHz then usually the minimum PSRR requirement is met out to 100-200kHz.
Personally, as a rough guide I would want to see good behaviour of the power supply up to 200kHz, do everything in your power to minimise the supply's impedance as seen by the appropriate parts leads up to at least that frequency. Remember, audio amplifiers start to fall apart above the audio band, but that's exactly where the FB mechanisms need to function correctly, and that's not going to be helped by the voltage rails glitching at those critical frequencies ...
I want to run on +/-85VDC rails for the sub (20Hz -50Hz) and woofers (50Hz - 180Hz twin 10 inch drivers 3.5R). I will be running 4 way active speakers that require 4 pairs of power amplifiers.
The Maida does not seem to be the best fit for me.
The TL783 is interesting. I don't know if it would and how well it would work with negative rail. I prefer the standard instead of LDO, as the output of the LDO is pretty much affected by capacitance.
Any other options?
The Maida does not seem to be the best fit for me.
The TL783 is interesting. I don't know if it would and how well it would work with negative rail. I prefer the standard instead of LDO, as the output of the LDO is pretty much affected by capacitance.
Any other options?
You can use the usual suspect (LM317T) to give you (say) 90V output voltage, just you have to take great care not to short the output. That's because this family of regs is floating. You probably can provide some sort of short circuit protection with judicious use of zeners.
In any case I'd suggest if you really, really need 500W then you look at delivering it with a bridged topology. This at a stroke halves your PSU impedance, nearly halves your total required PSU volts and also makes much better use of the available capacitors - akin to comparing half-wave with full-wave rectification from a PSU pov. Running bridged then allows you to use only one regulator, rather than needing two.
@twest - does anyone use those slow roll-off filters? Can't think of any advantage in selecting them myself My own DACs have a 50+dB brick wall filter around 17-18kHz
In any case I'd suggest if you really, really need 500W then you look at delivering it with a bridged topology. This at a stroke halves your PSU impedance, nearly halves your total required PSU volts and also makes much better use of the available capacitors - akin to comparing half-wave with full-wave rectification from a PSU pov. Running bridged then allows you to use only one regulator, rather than needing two.
@twest - does anyone use those slow roll-off filters? Can't think of any advantage in selecting them myself
My own DACs have a 50+dB brick wall filter around 17-18kHz
Last edited:
My experience is 317 based Maidas seem to have a tendency to hit SOA trouble, which is one of the reasons I didn't suggest a 337 based Maida for the negative rail. A pass device intended for HV use and suitable soft start addresses that. Though at only 90V SOA might not be too much of an issue, particularly with a 317HV---the blown 317 Maidas I'm aware of were on tube B+ supplies in the 300-500V range.
Bridging on 45V rails allows direct use of a 317HV but if you allow 15-25% margin for mains overvoltage events there's no turnkey integrated negative regulator I'm aware of for rails over 32-35V. Single supply off a regulator like 738 isn't a bad idea but makes managing startup and shutdown transients more involved. If a Maida's too much for NutNut I wouldn't particularly suggest going there. (Single supply also yields asymmetric headroom as the dropout's with respect to the positive rail which, if one's trying to maximize swing off the rails, isn't usually where one would want to have it.)
I think I may be a bit confused as to why one would care about a sub amp's performance at 20kHz, though.
Bridging on 45V rails allows direct use of a 317HV but if you allow 15-25% margin for mains overvoltage events there's no turnkey integrated negative regulator I'm aware of for rails over 32-35V. Single supply off a regulator like 738 isn't a bad idea but makes managing startup and shutdown transients more involved. If a Maida's too much for NutNut I wouldn't particularly suggest going there. (Single supply also yields asymmetric headroom as the dropout's with respect to the positive rail which, if one's trying to maximize swing off the rails, isn't usually where one would want to have it.)
I think I may be a bit confused as to why one would care about a sub amp's performance at 20kHz, though.
Last edited:
I was too quick to say that Maida did not seem like the one for me - I had a very quick look earlier at the schematic and I thought it was using an opamp, which would require a power supply for its own, hence would be a "NO" for me. From your later post I knew I was wrong. I had a second look and it seems that perhaps a 317/337 based maidas is an option.
But this can get very complicated because I will need to consider when the amplifier is turned on and off, each separate PSU (front end Maida and backend large capacitors) has different speed of charging and discharging which may cause problems to the amplifier.
But this can get very complicated because I will need to consider when the amplifier is turned on and off, each separate PSU (front end Maida and backend large capacitors) has different speed of charging and discharging which may cause problems to the amplifier.
Look more closely at how the op amp is powered when you have time. It's not any different than the error amp inside of a 317 or 337. Well, other than the complexity's more oviously exposed, but you'll have to deal with that and a variety of related sequencing problems one way or another to ensure SOA is not violated (welcome to power supply design ). You might also want to read the discussion of protection diodes in National's (er, TI's) LM317 datasheet.
). You might also want to read the discussion of protection diodes in National's (er, TI's) LM317 datasheet.I think I may be a bit confused as to why one would care about a sub amp's performance at 20kHz, though.
I have non-sub amps to build, but they still run on rail voltages of +/-56VDC.
For the sub and woofer amps, they will handle frequencies below 200Hz. But if they are noisy amps, would they pollute the higher frequency regions?
Look more closely at how the op amp is powered when you have time. It's not any different than the error amp inside of a 317 or 337. Well, other than the complexity's more oviously exposed, but you'll have to deal with that and a variety of related sequencing problems one way or another to ensure SOA is not violated (welcome to power supply design
I have built the lm317/337 regs many times in line level applications. The best sound is to use them as pre-regs then small LCR to get rid of the noise from the lm317/337 then CFP capacitor multiplier with huge, low impedance caps following it as in that case I could get electrolytic caps of any size and any impedance directly to the opamp pins without causing any LCR resonances whatsoever. If using lm317/337 alone then their output inductance of 400nH to 2uH would cause impedance peaks / ringings with low impedance caps, not the best possible sound with opamp circuits.
Could you measure a difference on the line outs between the above and a more typical output cap arrangement?
Possibly. Measure the sub's SPL and the amp's output noise and then you'll know.
I have returned from my holiday, during which I was thinking about the PSU.
I previously thought that passive LCR filter for the front end was not good enough hence was thinking about using a high voltage regulator instead. That was because in my standalone simulation of a front end LCR filter, using reasonable, practical value of capacitance the low freqency impedance 4 x 10R and 8 x 680uF = 5,440uF was at 8dB at 20Hz with a voltage drop of no more than 1.5V. In that simulation, the LCR filter makes ripples worse below around 55Hz, and is bearly useful below a few hundred Hz.
My power amp IPS and VAS draws 25mA. With 40R in series in the LCR this translates to 0.025 * 40 = 1V at DC.
However, due to the challenges I was facing to implement a good sounding high voltage regulator, I revisited the LCR filter approach. I was surprised to discover that the LCR filter is not as bad as I originally simulated.
When a diode is used in the front of the LCR filter to prevent "reverse" current and the LCR filter plugged into the real amplifier circuit with ripples on the rails (instead of a standslone LCR circuit with perfect voltage source), the diode works wonders! It buffers the voltage to prevent excessitve voltage drops and in addition smoothens the ripples by quite a lot!
The following are the simulation result of the PSRR of the front end LCR filter of D-C-RC-RC-RC-RCCCC (D=1N5819, R=10R, C=680uF):
20Hz: -26dB
120Hz: -38dB
2000Hz: -52dB
The rail PSU PSSR before the front end LCR filter:
20Hz: -11dB
120Hz: -24dB
2000Hz: -46dB
The IPS and VAS stages use CCS and cascodes which provide the following PSRR:
20Hz: -69dB
120Hz: -66dB
2000Hz: -56dB
Below are the combined PSRR:
20Hz: -107dB
120Hz: -128dB
2000Hz: -154dB
At higher frequencies up to 40,000kHz PSRR would not be much worse than the above figures, but the simulation took too long to complete and the ripples were too small for the simulation to be accurate therefore was not completed.
The price to pay are the bulk of the D-C-RC-RC-RC-RCCCC (D=1N5819, R=10R, C=680uF), as well as 1.5V voltage drop hence the loss of some power. The gain is the achievement of -107dB worst case PSRR.
The risk is free comparing to an active regulator.
Of course, a regulator will provide even much better PSRR but the risk of having possible higher fault rate (exceeding SOA of active devices, inrush current, turn on/off thumbs, reverse current, amplifier clipping under heavy load condition, etc) as well as the complexity do make me think trice before heading this path.
What do you think?
I previously thought that passive LCR filter for the front end was not good enough hence was thinking about using a high voltage regulator instead. That was because in my standalone simulation of a front end LCR filter, using reasonable, practical value of capacitance the low freqency impedance 4 x 10R and 8 x 680uF = 5,440uF was at 8dB at 20Hz with a voltage drop of no more than 1.5V. In that simulation, the LCR filter makes ripples worse below around 55Hz, and is bearly useful below a few hundred Hz.
My power amp IPS and VAS draws 25mA. With 40R in series in the LCR this translates to 0.025 * 40 = 1V at DC.
However, due to the challenges I was facing to implement a good sounding high voltage regulator, I revisited the LCR filter approach. I was surprised to discover that the LCR filter is not as bad as I originally simulated.
When a diode is used in the front of the LCR filter to prevent "reverse" current and the LCR filter plugged into the real amplifier circuit with ripples on the rails (instead of a standslone LCR circuit with perfect voltage source), the diode works wonders! It buffers the voltage to prevent excessitve voltage drops and in addition smoothens the ripples by quite a lot!
The following are the simulation result of the PSRR of the front end LCR filter of D-C-RC-RC-RC-RCCCC (D=1N5819, R=10R, C=680uF):
20Hz: -26dB
120Hz: -38dB
2000Hz: -52dB
The rail PSU PSSR before the front end LCR filter:
20Hz: -11dB
120Hz: -24dB
2000Hz: -46dB
The IPS and VAS stages use CCS and cascodes which provide the following PSRR:
20Hz: -69dB
120Hz: -66dB
2000Hz: -56dB
Below are the combined PSRR:
20Hz: -107dB
120Hz: -128dB
2000Hz: -154dB
At higher frequencies up to 40,000kHz PSRR would not be much worse than the above figures, but the simulation took too long to complete and the ripples were too small for the simulation to be accurate therefore was not completed.
The price to pay are the bulk of the D-C-RC-RC-RC-RCCCC (D=1N5819, R=10R, C=680uF), as well as 1.5V voltage drop hence the loss of some power. The gain is the achievement of -107dB worst case PSRR.
The risk is free comparing to an active regulator.
Of course, a regulator will provide even much better PSRR but the risk of having possible higher fault rate (exceeding SOA of active devices, inrush current, turn on/off thumbs, reverse current, amplifier clipping under heavy load condition, etc) as well as the complexity do make me think trice before heading this path.
What do you think?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.