dellama, to continue the conversation here - there is no need to accept compromises, all subjectively significant distortions can be removed from the picture if enough effort is applied.
The most important "job" is to remove pernicious low level distortions - these are, would be, very hard to measure by instrument; at the moment, learning to rely on one's ears I feel is the best approach.
Firstly, a square wave has a multitude of sine waves. In a perfect square wave, the sines go out to infinity. Good square wave response correlates with good frequency response and can also tell you a lot about the compensation design - minimal or no overshoot and fast settling even in the presence of capacitive load are indicators of good comp design.
Secondly, there is a lot of effort going on here to get higher OLG, and therefore LG on the CFA's.
If you increase the OLG, the phase accumulation increases, and you quickly get to the point where the comp design has to contend with the same issues you get in a VFA. I have designed both types, and the high loop gain CFA's have to use TPC, TMC or just plain MC to close the loop correctly to preserve stability.
But, to me, this approach loses some of the magic of a 'true' CFA: lower loop gains, very wide OLG etc. Sure, distortion is a bit higher, but you gain speed and bandwidth. My sx-Amp for example has a loop gain -3 dB of 60 kHz.
Why? Because the OPS pole on a classic low loop gain CFA falls at or close to the UGF ( open loop). The amp therefore requires very little compensation to preserve stability which would not be the case if the loop gsin was much higher. Without an input filter or an output inductor, the sx-Amp bandwidth is about 4.5 MHz ( 75 pF comp cap). Not good in practice since we really do not want this thing acting like an AM transmitter, but it gives some indication of the performance capability of low loop gain CFA.
And, just make it clear, I like both VFA and CFA.
Secondly, there is a lot of effort going on here to get higher OLG, and therefore LG on the CFA's.
If you increase the OLG, the phase accumulation increases, and you quickly get to the point where the comp design has to contend with the same issues you get in a VFA. I have designed both types, and the high loop gain CFA's have to use TPC, TMC or just plain MC to close the loop correctly to preserve stability.
But, to me, this approach loses some of the magic of a 'true' CFA: lower loop gains, very wide OLG etc. Sure, distortion is a bit higher, but you gain speed and bandwidth. My sx-Amp for example has a loop gain -3 dB of 60 kHz.
Why? Because the OPS pole on a classic low loop gain CFA falls at or close to the UGF ( open loop). The amp therefore requires very little compensation to preserve stability which would not be the case if the loop gsin was much higher. Without an input filter or an output inductor, the sx-Amp bandwidth is about 4.5 MHz ( 75 pF comp cap). Not good in practice since we really do not want this thing acting like an AM transmitter, but it gives some indication of the performance capability of low loop gain CFA.
And, just make it clear, I like both VFA and CFA.
Last edited:
Firstly, a square wave has a multitude of sine waves. In a perfect square wave, the sines go out to infinity. Good square wave response correlates with good frequency response and can also tell you a lot about the compensation design - minimal or no overshoot and fast settling even in the presence of capacitive load are indicators of good comp design.
Secondly, there is a lot of effort going on here to get higher OLG, and therefore LG on the CFA's.
If you increase the OLG, the phase accumulation increases, and you quickly get to the point where the comp design has to contend with the same issues you get in a VFA. I have designed both types, and the high loop gain CFA's have to use TPC, TMC or just plain MC to close the loop correctly to preserve stability.
But, to me, this approach loses some of the magic of a 'true' CFA: lower loop gains, very wide OLG etc. Sure, distortion is a bit higher, but you gain speed and bandwidth. My sx-Amp for example has a loop gain -3 dB of 60 kHz.
Why? Because the OPS pole on a classic low loop gain CFA falls at or close to the UGF ( open loop). The amp therefore requires very little compensation to preserve stability which would not be the case if the loop gsin was much higher. Without an input filter or an output inductor, the sx-Amp bandwidth is about 4.5 MHz ( 75 pF comp cap). Not good in practice since we really do not want this thing acting like an AM transmitter, but it gives some indication of the performance capability of low loop gain CFA.
And, just make it clear, I like both VFA and CFA.
I think this is a very good explanation.
Some caution needs to be applied when there is dependence on the OPS pole for compensation, since there are many things that can affect it, and there is really more than one pole in the OPS. We need to bear in mind that the ft of the output devices can depend on Ic and Vce, and in fact ft droop can be fairly bad in some devices at high current.
The key is that good PM and GM are maintained with a VFA or CFA over the full set of signal swings, load impedances (including reactive) and output currents, and over a good spread of output device characteristics. This is important to apples-apples comparisons (which are usually difficult anyway).
Cheers,
Bob
Alas, many less experienced designers see a 'good PM & GM with 8R load' as 'sufficient' for good stability. It's really just the first stage in stability design.
You need to check PM & GM with all possible (and some stupid) loads as well and do transient analysis with the same loads .. exactly as you should do in 'real life'.
Then you need to see if the amp is well behaved under all these conditions when overloaded too.
A good amp that behaves well on all these tests will have 'good PM & GM on 8R loads' .. but the converse is not guaranteed.
I like the 2 Ohm//2 uF 'torture test' and combos of anything lighter than this.
One of the reasons by the way I always use an output inductor - just 1uH is enough to allow any conceivable load to be driven, and often 0.5uH is enough. You can get away without the inductor - but you have to really lower the loop gain to get this to pass the torture test.
One of the reasons by the way I always use an output inductor - just 1uH is enough to allow any conceivable load to be driven, and often 0.5uH is enough. You can get away without the inductor - but you have to really lower the loop gain to get this to pass the torture test.
Actually this test is easy peasy for most amps.
It was devised to emulate a QUAD ELS-57 in the early 60's but doesn't look anywhere like it. It's mainly cos HiFi News used it in their tests of amps from that period that it remains in the mythology of amp testing.
For amps using the usual Zobel + Thiele network, the most difficult loads for stability are small capacitances .. from about 1n - 10n.
This is of course a stupid load. However, problems with these loads correlate well with instability with big guitar speakers on part of their waveform at certain frequencies, often triggered by mild (or severe) overload.
This is all 'real life' and I can't give a full theoretical explanation. However, many Golden Pinnae amps suffer from this so its easy to check.
With what frequencies or sweeps are you testing 2R//2uF ? because at 20K, that 2u has a Z of about 3.6 and goes lower with higher frequencies. I'd consider this not a torture test but an overload test and I don't expect perfect squares/sines with such load as several clamps may kick in (if you protected your design with clamps as needed).
I agree with this, I've experienced the same you describe. I'm now using a pure 22nF capacitance at the output after the inductor as you described, to provide for this minimum capacitance above the "crititical" range. It doesn't load the amp in a significant way anyways. It also ensures that PM/GM practically doesn't budge over the full capacitive range from 22n....2u
Last edited:
I test at 2.2 uF and 2 Ohms and then at loads lighter that this - all the way down to a few hundred pF. Coils and Zobel always used in my designs and I usually go for 60 deg PM minimum. If you use a 1 uH coil and 60 deg PM at 8 ohms, you are in the game for any conceivable speaker load.
Yes, the output coil resonated with capacitive load - but that's quite normal.
See sx-Amp write up for a discussion on how the comp for that amp was evolved.
I don't use normal current limiting on my amps. I prefer to overdrive the output and just use a fast over current trip where this makes sense.
Yes, the output coil resonated with capacitive load - but that's quite normal.
See sx-Amp write up for a discussion on how the comp for that amp was evolved.
I don't use normal current limiting on my amps. I prefer to overdrive the output and just use a fast over current trip where this makes sense.
Last edited:
This is akin to Prof. Cherry's recommended output network which I briefly mentioned in Dave Zan's "Middlebrook GFT probe" thread.
My own brief SPICE sims suggests this is a better way to go than the normal Zobel + Thiele that Self describes. However, I don't have 'real life' experience of this method. Your experience encourages me to look further.
Much of my poor SPICE efforts have been trying to replicate my Jurassic 'real life' experiences. I'm rather impressed by LTspice. It usually gives sensible results if you have good models like Bob Cordell's, and if you realise PCB tracks have evil Inductance & Capacitance too.
I'm loath to recommend stuff that I haven't tested in some form in 'real life'.
bonsai said:
After the initial design using analysis & PM/GM, I tend to use .TRANS to pick up problems and then switch back to PM/GM for the specific trouble cases. You can look at overload behaviour too.
I say more on my "TPC vs TMC vs pure Cherry" thread. Again I stress its all stuff you should do in 'real life' too. In SPICE world, you can do loadsa stuff like this without releasing the Holy Smoke.
For the big 'real life' guitar speaker, things seem to happen just above the bass resonance but before the impedance flattens off and becomes 'resistive'. My guess is the speaker impedance is falling fast and capacitive in this range.
Andrew, I'm really impressed by your protection scheme.
It's what I wanted to do in the 90's but was too expensive.
Last edited:
Bryston also uses a pure shunt capacitance at their output, but I think it is larger, on the order of 0.05uF.
I'm still a big believer in having a Zobel on the output node of the amplifier ahead of the L-R series network. This provides well-defined loading and damping for the EF output stage that can be physically very close to the output transistors, so that it is effective to very high frequencies, regardless of the wiring and other details of the L-R network or output capacitor.
Cheers,
Bob
This relies upon your confidence in the square wave being generated and in the instrument reading it. It also does not describe the simultaneous transient variations in level that all music has. Don't get me wrong, a good square wave is a good starting point but there are lots of caveats most people don't even consider. Has anyone found a sonic correlation with phase margin? In other words, if two amps are stable, will the amp with 85% phase margin sound better than one with 75%? The same questions for gain margin.Bonsai said:
Different amps react differently to gain and phase margin.
To even specify gain or phase margin assumes that they will be reduced during overload. In some amps they actually increase under overload. It depends on exactly which part gets overloaded and how it relates to the loop.
So I wouldn't hang my hat on either of them, except when dealing with topologies that have a known overload behavior.
To even specify gain or phase margin assumes that they will be reduced during overload. In some amps they actually increase under overload. It depends on exactly which part gets overloaded and how it relates to the loop.
So I wouldn't hang my hat on either of them, except when dealing with topologies that have a known overload behavior.
What about a sustained short-circuit? (long enough to blow OPS power fuses). Don't you want VAS currents or BE junction currents to be limited? I'm really trying to make my amp be able to handle a sustained short-circuit, even after the OPS fuses have blown. I'd hate to lose expensive parts and / or PCB tracks...
My comment about square waves was referring to the theoretical case. The rise time of a real world generator is not 0, so the sine harmonics of course will not extend out to infinity.
See Self et al for a discussion on why music signals are not a special case - an amp can be fully characterized from just sine and square wave testing.
Amplifier loop gain and ULGF are affected due to device parameters shifting as the voltage and or current applied to them changes - the VAS Cob is a good example. Sine plus a small superimposed square wave helps detect this type of problem so they are also not unique to music signals.
