Bob Cordell's Power amplifier book

Hi Bob!
I am thinking og reviving your 50W HEXFET amplifier project that I started 30+ years ago. I did the layout of the boards, bought components, even built the box and had a custom ordered transformer. But I never came to actually producing the PCBs. So my question is this: Is it worthwhile in your opinion, to make this amp now? Will it compare favourably to the DH-220C? Which btw,I already have made a PCB for, but not started building.

A second question about output transistors: In my schematic I have doubled the output tranistors. Is this possible? To balance the transistors, maybe I would need to add source resistors to the HEXFETS?

My original PCB layout and box:

View attachment 1453599

View attachment 1453600

Today's schematic:

View attachment 1453601
Thanks for your interest in my Mosfet amplifier with error correction. It looks like you have done a nice job with your initial work on it. I should note that the original JAES article describing it in detail is available on my website at cordellaudio dot com. The well-reviewed Halcro amplifier line uses a slightly revised version of that circuit, and it is described in their patent 5,892,398 issued 4/6/1999. A high-powered pro amplifier also used that circuit in the late 1980s. I also built a prototype of a 200 wpc version of the circuit, but never fully finished it due to job commitments.

The circuit you have shown will perform well is is worth completing, but with a couple of suggestions. A higher-power version is recommended, at least 100 Watts. The design is quite scalable as long as transistors with suitable breakdown voltages are used. The types of transistors I use in the examples in my amplifier book are suitable.

I designed and built two other amplifiers with this circuit, but without the error correction using vertical MOSFETs IRF240 and IRFP9240). One used a single output pair and was rated for 125 watts. Four of them were used in each channel of the self-powered active Athena loudspeakers I built. The other was a 200-watt amplifier that used two pairs of the HEXFETs.

The circuit with error correction does sacrifice some output voltage because of the needed extra headroom for the error correction circuitry and the higher turn-on voltage of vertical MOSFETs as compared to BJT output transistors. That is why boosted supply rails were used in the original design. The use of Baker clamps and cascoded VAS also required additional headroom. The required current for the rail-boost supply is low, but this is still a cost-adder. Even conventional power amplifiers benefit from boosted rails, although not boosted as much. A small transformer with dual secondaries and small bridge rectifiers providing a voltage that can ride on top of the main rails is one simple way to achieve the boost. This is part of the reason why it makes sense to build a version of this amplifier that is capable of at least 100 watts.

Using two or more pairs of output MOSFETs parallel-connected is not a big problem as long as some matching of the pairs being paralleled is done. This is discussed in the original article. Often, a pair that comes out of the same tube will be adequately matched. Source resistors do very little to mitigate the effects of MOSFET mismatch, so they are often not used in many MOSFET designs. However, I routinely use 0.22 or 0.39-ohm source resistors in MOSFET designs if for no other reason than to monitor bias currents. However, due to the inherent differences in transconductances between N-channel and P-channel mosfets when operated at the same bias current, I have found that using smaller source resistors for the P-channel devices than for the N-channel MOSFETs can reduce distortion by a substantial amount. I use 3-watt Metal Oxide Film resistors for source or emitter resistors in my designs (paralleled if necessary).

The lateral-MOSFET DH-220C amplifier has been described as good-sounding by those who have built it. It is also scalable to higher power in the professional P230 Hafler and the DH-500 Hafter. It measured distortion is of course somewhat higher that the amplifier with error correction. Depending on how far along you are on each design, you might want to finish the DH220C amplifier first.

Cheers,
Bob
 
Thanks a lot for your time and in-depht explanations, Bob!
Do I understand correctly, that by keeping the two output pairs and raising the supply voltage, this amp is good for 100W?

Made a quick change on the schematic, including source resistors. Unsure if the difference in values are ok?

Cordell_100W_skjema_mg.JPG


About going for DH220C... I could do that, since the PCB is alread made. It is just that this PCB is one that I am not satisfied with. Too big, too much waste space. Guess I was in a hurry. Here are the 50W (now upgraded to 100W) PCB (not produced) and the DH220C driver and servo PCBs (produced).

Cordell_100W_PCB_mg.JPG


Cordell_DH-220C_driver_layout_mg.JPG


Servo

Cordell_DH-220C_complete_servo_mg_layout.JPG
 
  • Like
Reactions: hbtaudio
Yes, you are basically correct. However, a single output pair is good for at least 100 watts, as long as the supply voltage is adequate and there is an appropriate amount of heat sinking. If you use two output pairs you can achieve somewhat lower distortion due to the higher transconductance available from two pair, each pair being operated at the same bias current as with a single pair (in other words, double the total bias current). In addition, of course, with two pairs it is safe to go up to higher power, perhaps 200 Watts. Also, using two pairs does not obligate you to use higher total bias current.

Unlike BJT output stages wanting to conform to the Oliver criteria for bias current to achieve low crossover distortion, MOSFETs are more tolerant of different values of bias current. For them, to first order, the more bias current the better.

I should mention that the original MOSFET EC amplifier used an NPD5564 dual monolithic low-noise JFET. They are no long available AFIK. The LSK389 from Linear Systems is a good substitute with very low noise.

Cheers,
Bob
 
  • Like
Reactions: hbtaudio and m0rten
Thanks again!
Then it will be 2 output pairs. It is easier to drop a pair than to add a pair...
Since you did not comment on my source resistor values, I guess they are ok enough to try.
I do have the 2N5564 which probably is equivalent to NPD5564. I could use this. Is the 2SK146 an absolute "NO", or is it just noisier?
Borbely used these quite often in the past. Even in the EB-1195/221 MC/MM hybrid RIAA preamp... Are they really that noisy?
You really did help revive this old projekt!
Maybe the DH220C will have another go with a new layout later.

😎
 
Yes, you are basically correct. However, a single output pair is good for at least 100 watts, as long as the supply voltage is adequate and there is an appropriate amount of heat sinking. If you use two output pairs you can achieve somewhat lower distortion due to the higher transconductance available from two pair, each pair being operated at the same bias current as with a single pair (in other words, double the total bias current). In addition, of course, with two pairs it is safe to go up to higher power, perhaps 200 Watts. Also, using two pairs does not obligate you to use higher total bias current.

Unlike BJT output stages wanting to conform to the Oliver criteria for bias current to achieve low crossover distortion, MOSFETs are more tolerant of different values of bias current. For them, to first order, the more bias current the better.

I should mention that the original MOSFET EC amplifier used an NPD5564 dual monolithic low-noise JFET. They are no long available AFIK. The LSK389 from Linear Systems is a good substitute with very low noise.

Cheers,
Bob
I forgot one thing: Which transistors may/should or should not be fitted on the main heatsink?
At the moment the drivers (Q26, Q27) and bias transistora Q22, Q23 are all on the edge with the HEXFETS.
In the article you mention Q22 to be mounted on the main heatsink, so I suppose that temco for the NPN/PNP bias pair work in opposite direction, meaning Q23 is NOT to be put on the heatsink.
Is that right?
 
Hi Freecrowder!
If you want, I will send you a set of PCBs for the DH220C driver/servo if you PM you name/address. I cannot guarantee that there will be a new version, so you should say yes
If you want to play with my KiCAD project, here is the KiCAD project folder including Gerbers. Hope you figure it out.
Anyone using this KiCAD project should note that there is an error in the produced Servo PCB. (And Gerbers and schematic, yes)
It is small but obvious, but I will tell anyway.
The OP AMP input diodes should connect to ground on one side, They don't.
Easy fix, and now you know!
 
  • Like
Reactions: anatech
I forgot one thing: Which transistors may/should or should not be fitted on the main heatsink?
At the moment the drivers (Q26, Q27) and bias transistora Q22, Q23 are all on the edge with the HEXFETS.
In the article you mention Q22 to be mounted on the main heatsink, so I suppose that temco for the NPN/PNP bias pair work in opposite direction, meaning Q23 is NOT to be put on the heatsink.
Is that right?
The source resistors you have chosen look fine. If I read it properly, you chose 0.39 ohms for the N-channel and 0.22 ohms for the P-channel MOSFETs. As discussed earlier, these are relatively non-critical and do not reduce the need to do some checking of matching between MOSFETs of the same polarity. If you have a sensitive distortion analyzer, you can try shorting out the P-channel source resistors to see if distortion goes down. However, even at 20 kHz, distortion will be difficult to measure if the EC is working properly along with the rest of the circuitry. Similarly, adjusting the EC distortion trim pot may be difficult since the closed loop distortion will be quite low. If you probe the distortion at the emitter follower buffer at the output of the VAS under closed-loop conditions, you may be able to see the quasi-open-loop distortion of the output stage, since that signal will be the error signal fed to the output stage that is required to make the final output have very low closed-loop distortion.

To check matching of MOSFETs, look at the gate-source voltage when the MOSFET is operating at 100 mA with Vds of 15V or more in a simple test setup. No heat sink is required for this test if you don't take too long to do the test. Just vary the gate-source voltage with a variable power supply source until measured drain current is 100 mA. Put a 1-k resistor in series with the gate to ensure no oscillation. Put a 0.22-ohm resistor in series with the source and look for 22 mV across it to operate the MOSFET at 100 mA. Matching within 10% of Vgs is probably adequate, and is not super-critical.

IFIK, many 2N5564 dual transistors were made with two die in the package. I'm not familiar with the 2SK146. The LSK389 is probably the best choice. If you contact Linear Systems you may be able to get a couple of samples. Make sure you review the datasheet to assure proper pinout.

Only one of the error correction transistors went on the heat sink in the prototype I did. Its wire length should be fairly short so as not to degrade its speed in carrying out the error correction. In my original implementation I had the drivers on the circuit board with their own miniature heat sinks, but it probably works OK if they are on the heat sink since they will warm up anyway in either case. Do evaluate the thermal stability in operation, especially to avoid any tendency to thermal runaway. Turn it on cold with the bias set low, then increase the bias to the nominal value and watch the bias as things warm up and adjust as necessary. Give it plenty of time to heat up the heat sink to a stable value while making minor adjustments as it does so. Then allow the amplifier to cool off for a long time. Then turn it on again and see what the bias does as it starts from cold and warms up. If more bias thermal compensation is needed, you can try putting both of the EC transistors on the heat sink. The actual bias current in the output stage is not critical, but is best at around 100-200 mA per MOSFET.

Watch for signs of parasitic output stage oscillations with a wideband scope for various signal conditions, including near clipping with and without an 8-ohm load. Vertical MOSFETs are very fast and like to oscillate, as described in my JAES paper.

Cheers,
Bob
 
I can't believe you take the time to give all this time and valuable information to an amateur like me!
All my questions adressed and more!

On matching MOSFETS, I made this device some time ago while building the PASS Aleph 5.
Not very sophisticated, but easy to use. Hope this will do the job.

Thanks again. I will try to see if I can do a better layout, as I had to accept a couple of vias on the PCB. I don't like vias as they tell me I am a lousy router...

😎 morten


PASS_FET_tester_skjema.jpg


PASS_FET_tester_module.jpg



Matching for Aleph 5

PASS_FET_tester_IRFP240.jpg


Last PCB including vias...


Cordell_100W_PCB_mg.JPG
 
If I read it properly, you chose 0.39 ohms for the N-channel and 0.22 ohms for the P-channel MOSFETs. As discussed earlier, these are relatively non-critical and do not reduce the need to do some checking of matching between MOSFETs of the same polarity. If you have a sensitive distortion analyzer, you can try shorting out the P-channel source resistors to see if distortion goes down.
Agree, the Error Correction watches the "error" from the output node that is after the source resistors. This gives the advantage over something like CFP, which only does its "Error Correction" before the source resistors for the sake of thermal stability. Thus, the source resistor value doesn't really matter much here.

BTW, in my simulation, with EC topology, there is an optimal bias point for lowest distortion, somewhere between 50mA to 80mA. Basically, EC increases the gm of the output stage by a lot, which makes it more like CFP. Higher bias could improve gm further, but it diminishes the return, and only widens the crossover region, which makes distortion worse.
 
  • Like
Reactions: hbtaudio
... I do have the 2N5564 which probably is equivalent to NPD5564. I could use this. Is the 2SK146 an absolute "NO", or is it just noisier?
Borbely used these quite often in the past. Even in the EB-1195/221 MC/MM hybrid RIAA preamp... Are they really that noisy? ...
2SK146 is low noise, high GM part used primarily in preamps and LONG discontinued. Although I know the old Acoustat TNT-200 and TNT-120 power amps used them as front-end diff amps. Higher gm (than NPD55640 means you may have to rework the compensation scheme due to gain change. The cost for higher gm and lower noise (you NEVER get something for nothing) is larger device capacitance, but your cascoding will help mitigate a lot of that. If you really don't need to save them for use in a preamp, go ahead and use them for your power amp. Then you can brag about having a very low noise front end! 😉
 
  • Like
Reactions: hbtaudio and m0rten