Audio Power Amplifier Design book- Douglas Self wants your opinions

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Here is something I would like opinions on. Looking at the Harmon Kardon Citation 12 it resembles the Naim NAP160 a little. Neither look exactly like the RCA which is claimed as the Genesis circuit. Myself LM741 seems more relevant ( Circa 1963 and emerging 1968 ). I would say LM741 and this RCA circuit along with the Sinclair Z30/50 are the mythical originator designs of the Blameless amplifier. Gogny of 1967 is not far off also. It's major difference being the feedback is summed at the input like an H C Lin design. It has a long tail pair input and VAS like a blameless amp ( no CCS or bootstrap! ). It gives 50 watts into 1 R for a ribbon driver and uses 2N3055 in a Darlington as at that current the gain is about 5 when pushed into 1R. It is said this form of feedback is better. Looking at the Gogny we still could do that.

The side issue is the use of MOS FET's. It seems a long time ago there were versions of the H K that used the cheaper MOS FET power devices, I suspect only in prototype. This seems to be slightly before the Hitachi version. H C Lin has a MOS FET patent about 1968 that looks to be a usable device. I remember Julian Vereker of Naim Audio asking me in about 1977 if I knew of a source of them so I guess he read the same texts?

My question is . Does the MOSFET gate capacitance form a useful pole and assist the amplifier? It is often stated as the reason to strongly distrust MOS FET's . My thinking is 100 watts 20 kHz is an unlikely musical requirement. My own experience with the Exicon devices is 6 mA or slightly lower is optimum VAS current ( 5.6 mA ) , Cg is about 700 pF each. Perhaps the reason is the amp will offer 5 watts with ease and then start to give up if excessive HF is required ( graphs seem to contradict that and resistor loads). Could it be we sometimes do things at HF that briefly ask for more ? Fault conditions or reactive load demands? What I might like about the sound is the amplifier seems to self limits at the critical moment? A friend who built 1000 watt versions said the sound only got better as FET's were added . That was in bridge with still only 7 mA VAS. I think that was 16 devices total. Could it be lower Ron and higher Cg that where having the desired effects? The Hitachi from memory is about 22 V/uS ( I think it calculates a bit higher). I never saw or heard anything that seemed like slew limiting with it. The input is hardly filtered ( 2K2 47 pF ) and no feedback loop compensation. The distortion spec of the Hitachi is very impressive at > 30 kHz. Surprisingly power output is maintained also. It is said the 2 SD756 and 2SB716 transistors are part of that. There are only 5 transistors , 2 FET's and 1 diode in the circuit. The design will easily take two extra FET's ( 200 watts 4R ) . Bias is by resistor and is fine ( 20 to 100 mA depending on taste ,100 mA recommended per pair , less sounds more like bipolar, even 5 mA starts to correct HF distortion on an analyzer ). I never calculated the loop gain of the Hitachi, As the second LTP is degenerated by 100 R it looks to be modest. The input pair at 1 mA total with no input pair current mirror. All a bit of a mystery. Conversely why are bipolar amps so fussy by comparison? Could it be the speed of the MOS FET's in the real situation is so much faster ? As Exicon never say much about it I have always assumed it no better than high grade bipolar's. The Audio FET is not much like industrial FET's. Typically I need less to bias them than bipolar's ( 1.2 V is usually enough for the pair, 170 R at 7 mA ).

Looking at valve amps HF square-wave output is not great. The Hitachi will easily out perform that ( a factor of 10 I would say if what is typically built these days ). The question is how can the Hitachi brake all of the rules and yet show no major defect? If you have never heard one I supect if it was introduced into a blind test you would like it.
 
Nigel,

Some of your questions are answered in this application hints from Hitachi.

Hope this helps
 

Attachments

  • H-1.jpg
    H-1.jpg
    57.4 KB · Views: 414
  • H-2.jpg
    H-2.jpg
    71.9 KB · Views: 404
  • H-3.jpg
    H-3.jpg
    68.4 KB · Views: 403
  • H-4.jpg
    H-4.jpg
    62.5 KB · Views: 399
  • H-5.jpg
    H-5.jpg
    54.9 KB · Views: 413
This is very useful, I only ever saw the last page before. It makes the mystery deepen as the trans-conductance is seemingly very poor. It does beg the questions, how is it with all this against it the MOS FET seems to perform at least as well as any bipolar stage allowing for Ron.

Reducing the VAS current to 5.6 mA the square-wave performance starts to go off at 30 kHz. The sound to my ears is optimum when Exicon at that level. I suspect the small HF loss is an advantage.

The 12 K and 8,8 nF is placed to help the double VAS balance at > 30 kHz. It can be tuned. You should find the double VAS balances nicely withing 2% on the input pair. Not bad for a crude circuit. I plan to try a current mirror to replace the 5K1 ( 3K9 on mine ). That could be BCV61 ( 62 is the other polarity ). It is only a 20 V device. The 2 V clamping of VAS will protect it. I doubt you will better BCV 61 if using discreet parts. I have a feeling it will do little of advantage.

A tail CCS is an OK idea (to replace 62 K ) . Works almost as well without it. 220 R seems an ideal gate stopper. As far as I know it is the inductance that a cheap resistor has that is working. If you are as stupid as me you will find the gate oscillates all the time if measured ( not recommended ). It seems not to emerge from the source but can circle back through the VAS. With 220 R it is gone. Still there at the gate if looking. The wave is about 5 MHz on the one I tested and a very pure sine wave so it seemed. Hard to say because the scope is tailing off at 5 MHz. I didn't have my fancy scope then.

If you look very carefully the 100 kHz distortion it is 0.005 % at 30 watts. Hard to beat that if any amp and if you did to what advantage? Less components than any op amp with far higher performance.

Thanks Tommy.
 
The 12 K and 8,8 nF is placed to help the double VAS balance at > 30 kHz.

The role of R12 is to keep the thermal dissipation in Q4 the same as in Q5.
Some designs replace it with a common base transistor with its base connected to ground.
This common base transitor is sometimes shunt at high frequencies by a capacitor between emitter and collector.

Quite strange is the "Miller" cap of 15 pF.
I think a Samuel Groner's advice for this kind of topology (usually named "Hitachi") is to connect this cap from base of Q4 to ground and to add another one, of same value, between base and collector of Q5.
 
This is very useful, I only ever saw the last page before. It makes the mystery deepen as the trans-conductance is seemingly very poor. It does beg the questions, how is it with all this against it the MOS FET seems to perform at least as well as any bipolar stage allowing for Ron.

Reducing the VAS current to 5.6 mA the square-wave performance starts to go off at 30 kHz. The sound to my ears is optimum when Exicon at that level. I suspect the small HF loss is an advantage.

The 12 K and 8,8 nF is placed to help the double VAS balance at > 30 kHz. It can be tuned. You should find the double VAS balances nicely withing 2% on the input pair. Not bad for a crude circuit. I plan to try a current mirror to replace the 5K1 ( 3K9 on mine ). That could be BCV61 ( 62 is the other polarity ). It is only a 20 V device. The 2 V clamping of VAS will protect it. I doubt you will better BCV 61 if using discreet parts. I have a feeling it will do little of advantage.

A tail CCS is an OK idea (to replace 62 K ) . Works almost as well without it. 220 R seems an ideal gate stopper. As far as I know it is the inductance that a cheap resistor has that is working. If you are as stupid as me you will find the gate oscillates all the time if measured ( not recommended ). It seems not to emerge from the source but can circle back through the VAS. With 220 R it is gone. Still there at the gate if looking. The wave is about 5 MHz on the one I tested and a very pure sine wave so it seemed. Hard to say because the scope is tailing off at 5 MHz. I didn't have my fancy scope then.

If you look very carefully the 100 kHz distortion it is 0.005 % at 30 watts. Hard to beat that if any amp and if you did to what advantage? Less components than any op amp with far higher performance.

Thanks Tommy.

Hi Tommy,

First, some of the negativity you may hear about the MOSFET is born of ignorance and prejudice. Each type of device, be it BJT, Lateral MOSFET or Vertical MOSFET, has its own advantages and challenges. These are outlined in my book in a comparison that I believe is fair and honest.

Transconductance for MOSFETs, and especially Lateral MOSFETs is lower than that of BJTs, leading to what I call transconductance droop. This leads to greater low-order crossover distortion, and to the condition that MOSFETs like to be biased somewhat higher to get higher transconductance. MOSFETs do not really have an optimum bias like BJTs do - the more the better within reason and dissipation limits. This means that you are free to get a bigger class A region. It also means that MOSFET amplifiers are less prone to thermally-generated bias errors. You will never have your MOSFET amplifier become bias-starved under dynamic conditions.

I prefer Vertical MOSFETs because they achieve higher transconductance and result in less transconductance droop. They are excellent devices for audio as long as properly designed and driven. Verticals are what I used in my MOSFET power amplifier with error correction, which achieved THD+N below 0.001% at all power levels back in 1983.

Of course, MOSFETs are generally faster than BJT output transistors.

Bottom line is that a MOSFET output stage will have greater static crossover distortion all else remaining equal, than a properly designed BJT output stage when the BJT is truly operating at its optimum bias (which may often be only on the bench when playing a sinusoid in steady state). If you are looking for the very best on-the-bench or simulated THD, without resort to error correction, go with BJTs. If you go with BJTs, go with the ThermalTraks for better bias tracking. I have had good results with the ThermalTraks.

A 10C dynamic change in BJT die temperature will change Vbe by about 22mV, which is significant in comparison to the 26mV optimum bias across RE in a BJT. This is a great oversimplification, and things are not quite as bad as this suggests, but you get the picture. It is very easy to get a 10C die temperature swing that will not be compensated by the usual bias temperature compensation quickly enough. This is why ThermalTraks are preferred.

My Vertical MOSFET amps achieve around 0.01 to 0.02% THD or lower at 20kHz at any power level up to rated full power. This is with reasonably conservative compensation with gain crossover frequency at about 1MHz.

That Hitachi amplifier is not a very good design, and you can do much better.

Cheers,
Bob
 
Hello
Sorry for small off-topic, but I am looking for some advice/explenation.
I am going to try CFP or darlington arangement of LTP input stage with current mirror.
It looks that CFP has a higher gain but I do not know which configuration is better from the technical and practical point of view (I have never tryed it before). Both have very high overal gain and in/out impedance but soundwise I have no clue.
I would be very happy If someone can write a few simple words to compare this two input stages.

BIG thanks.
Peter
 

Attachments

  • input ltp.png
    input ltp.png
    42.1 KB · Views: 183
AX tech editor
Joined 2002
Paid Member
Hi Tommy,

First, some of the negativity you may hear about the MOSFET is born of ignorance and prejudice. Each type of device, be it BJT, Lateral MOSFET or Vertical MOSFET, has its own advantages and challenges. These are outlined in my book in a comparison that I believe is fair and honest.

Transconductance for MOSFETs, and especially Lateral MOSFETs is lower than that of BJTs, leading to what I call transconductance droop. This leads to greater low-order crossover distortion, and to the condition that MOSFETs like to be biased somewhat higher to get higher transconductance. MOSFETs do not really have an optimum bias like BJTs do - the more the better within reason and dissipation limits. This means that you are free to get a bigger class A region. It also means that MOSFET amplifiers are less prone to thermally-generated bias errors. You will never have your MOSFET amplifier become bias-starved under dynamic conditions.

I prefer Vertical MOSFETs because they achieve higher transconductance and result in less transconductance droop. They are excellent devices for audio as long as properly designed and driven. Verticals are what I used in my MOSFET power amplifier with error correction, which achieved THD+N below 0.001% at all power levels back in 1983.

Of course, MOSFETs are generally faster than BJT output transistors.

Bottom line is that a MOSFET output stage will have greater static crossover distortion all else remaining equal, than a properly designed BJT output stage when the BJT is truly operating at its optimum bias (which may often be only on the bench when playing a sinusoid in steady state). If you are looking for the very best on-the-bench or simulated THD, without resort to error correction, go with BJTs. If you go with BJTs, go with the ThermalTraks for better bias tracking. I have had good results with the ThermalTraks.

A 10C dynamic change in BJT die temperature will change Vbe by about 22mV, which is significant in comparison to the 26mV optimum bias across RE in a BJT. This is a great oversimplification, and things are not quite as bad as this suggests, but you get the picture. It is very easy to get a 10C die temperature swing that will not be compensated by the usual bias temperature compensation quickly enough. This is why ThermalTraks are preferred.

My Vertical MOSFET amps achieve around 0.01 to 0.02% THD or lower at 20kHz at any power level up to rated full power. This is with reasonably conservative compensation with gain crossover frequency at about 1MHz.

That Hitachi amplifier is not a very good design, and you can do much better.

Cheers,
Bob

Bob, thanks for that review! One question: when you say that optimum bias with MOSFETs is less critical, what is your view re: Gm doubling? Is there not a tradeoff between bias setting for low xover and Gm doubling?

Jan
 
The role of R12 is to keep the thermal dissipation in Q4 the same as in Q5.
Some designs replace it with a common base transistor with its base connected to ground.
This common base transistor is sometimes shunt at high frequencies by a capacitor between emitter and collector.

Quite strange is the "Miller" cap of 15 pF.
I think a Samuel Groner's advice for this kind of topology (usually named "Hitachi") is to connect this cap from base of Q4 to ground and to add another one, of same value, between base and collector of Q5.

If you change the 12 K as you say it shifts things around. Even on an old Ferrograph distortion test set it was noticeable that it was mimicking the VAS loading of the other side. I did speculate about building a bridge amp using the two VAS sides. Never really worked out how best to do the feedback.

My amp uses 2 x 27 pf . Looks to me that Hitachi only though side one to be required (single 15 pF ).

Instead of ground I have tried taking Cdom to the output sources via 100 R and a split capacitance ( used the Self split Cdom and 2K2 and worked from there). I converted to single VAS to make it easier. I think it did do something and gave no obvious problems. The bigger problem is I need a new oscillator to measure these things properly. Laziness and other things seem to always put the day off. What I did measure was about 10 dB improvement at 50 kHz using this idea. This wasn't quite as good as a double VAS without it which almost is at the limit of my test gear. When added to a Complimentary feedback pair I measured no obvious improvement. This proves the MOS FET to have a disadvantage, however one that can be overcome. My benchmark is - 80 dB 50 kHz 10 watts.
 
Hello
Sorry for small off-topic, but I am looking for some advice/explenation.
I am going to try CFP or darlington arangement of LTP input stage with current mirror.
It looks that CFP has a higher gain but I do not know which configuration is better from the technical and practical point of view (I have never tryed it before). Both have very high overal gain and in/out impedance but soundwise I have no clue.
I would be very happy If someone can write a few simple words to compare this two input stages.

BIG thanks.
Peter

In a roundabout way this answers your question. He manages to make a strong argument for a complimentary pair single input ( not long tail pair ). Many facts and figures so good reading.

MJR7-Mk5 Mosfet Power Amplifier
 
Bob, thanks for that review! One question: when you say that optimum bias with MOSFETs is less critical, what is your view re: Gm doubling? Is there not a tradeoff between bias setting for low xover and Gm doubling?

Jan

Hi Jan,

With MOSFETs it is virtually impossible to get gm doubling because of the lower transconductance and the transconductance characteristics of the devices.

Recall that gm doubling occurs in BJT output stages in the central region of the output current swing, where both the top and bottom transistors are contributing transconductance. When the magnitude of the output current causes the stage to exit the small class A region of the class AB output stage, one of the transistors turns off and no longer contributes transconductance. If the idle bias is set high, the gm of the transistors can be significantly higher than 1/RE. In the limit, in the central region, the gm of the stage is 2 * 1/RE. Outside the class A region, the gm of the stage is 1/RE, since only one of the RE resistors is in play. Bottom line is that an over-biased BJT output stage will have too much gm in the central class A region.

The inverse is generally true for MOSFETs. They have what I coined to be "transconductance droop" in the central region. Biasing the MOSFETs at a higher current increases their gm and reduces the droop, but almost never enough to eliminate the droop. You can bias them as hot as you like without risk of gm doubling.

Cheers,
Bob
 
Bob can you answer the mystery of the low trans-conductance and yet excellent distortion of the Hitachi design? If we take a quad of FET each to have 700 pF CISS that looks impossible for a 8 mA VAS. Hitachi show the distortion graph to be almost as good at 100 kHz as 1 kHz. This is especially true if thinking 1 watt 20 kHz for a tweeter is a sensible limit.

The FET output pair is said to have about 0.8% THD and a complimentary feedback pair would be 20dB better. How come the 5 transistors of the Hitachi could realize 200 watts 4 R with miniscule distortion? Not least when 8 mA of VAS looks to be incapable of driving the FET's CISS? Is it that something undisclosed is badly wrong with bipolar transistor amps ? Seems to me all the slewing talk is telling that story. I think the Hitachi is a bout 20 V/us . Never had any reason to think it couldn't cope. Could it be the switch nature that the FET doesn't have is a real problem. Could be the bias is not maintained under signal conditions? I have seen chunks out of waves like crossover distortion usually higher up the wave. I take this to be TID.
 
Bob can you answer the mystery of the low trans-conductance and yet excellent distortion of the Hitachi design? If we take a quad of FET each to have 700 pF CISS that looks impossible for a 8 mA VAS. Hitachi show the distortion graph to be almost as good at 100 kHz as 1 kHz. This is especially true if thinking 1 watt 20 kHz for a tweeter is a sensible limit.

The FET output pair is said to have about 0.8% THD and a complimentary feedback pair would be 20dB better. How come the 5 transistors of the Hitachi could realize 200 watts 4 R with miniscule distortion? Not least when 8 mA of VAS looks to be incapable of driving the FET's CISS? Is it that something undisclosed is badly wrong with bipolar transistor amps ? Seems to me all the slewing talk is telling that story. I think the Hitachi is a bout 20 V/us . Never had any reason to think it couldn't cope. Could it be the switch nature that the FET doesn't have is a real problem. Could be the bias is not maintained under signal conditions? I have seen chunks out of waves like crossover distortion usually higher up the wave. I take this to be TID.

Hi Nigel,

I assume you are referring to the amplifier in post #2322. I honestly do not see any way that the Hitachi design can achieve 0.01% THD at 100kHz at 100W into a 8-ohm load. The input pair is undegenerated and the output MOSFETs are driven with no buffers from the VAS. I have not simulated it, but I am bewildered by their claims.

BTW, I know of no THD analyzer that can give even remotely accurate THD readings at 100 kHz, since one would want at least a 500kHz residual bandwidth to capture even the first few harmonics. I think many THD analyzers often use an 80kHz residual bandwidth. My original THD analyzer used a 200kHz bandwidth for THD-20 measurements, but even that would be insufficient for a THD-100 measurement.

It is, of course possible in principle that Hitachi used an RF spectrum analyzer to look at the harmonics, but at the time I'm not sure any such analyzers had enough dynamic range to see the distortion lines with any accuracy (one would want 100dB of dynamic range).

Cheers,
Bob
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.