Before I can compare anything, I need to get a reliable baseline of the circuit that SuzyJ presented and unfortunately I can't seem to get less than about 0.01% THD so there is something up with my sim and or the models I have used.
With the specified components and values in the original schematic, the output stage quiescent current is close to 0.
When I replace THE MJE340/350 transistors with KSA1381 and KSC3503 in the sim then the output stage will draw about 50mA and the THD drops to 0.005% which leads me to the question of why are the MJE transistors specified for Q9 and Q10, surely the KSA and KSC can be excellent CCSs or is there another reason I have overlooked?
Sorry if this has been answered previously.
@Cyberpup, I have the original AEM6000 amps and haven't simulated any SuzyJ designs, this is the first one and only just done today but thanks for drawing my attention to SuzyJ's other threads, I need to get up to speed with those and will continue with this sim and will also compare JFE2140 with other JFETs 🙂
For now, here is an Ltspice sim that functions but needs more work, note the input cap is intentionally shorted to prevent artificially inflated THD.
The models are included in the sim.
With the specified components and values in the original schematic, the output stage quiescent current is close to 0.
When I replace THE MJE340/350 transistors with KSA1381 and KSC3503 in the sim then the output stage will draw about 50mA and the THD drops to 0.005% which leads me to the question of why are the MJE transistors specified for Q9 and Q10, surely the KSA and KSC can be excellent CCSs or is there another reason I have overlooked?
Sorry if this has been answered previously.
@Cyberpup, I have the original AEM6000 amps and haven't simulated any SuzyJ designs, this is the first one and only just done today but thanks for drawing my attention to SuzyJ's other threads, I need to get up to speed with those and will continue with this sim and will also compare JFE2140 with other JFETs 🙂
For now, here is an Ltspice sim that functions but needs more work, note the input cap is intentionally shorted to prevent artificially inflated THD.
The models are included in the sim.
My latest amps based on this topology are here: https://www.diyaudio.com/community/threads/a-small-class-h-lateral-mosfet-amplifier.365176/
If I disable the class H switching by just running both rails at the same voltage, they do single digit ppm thd. Most of the improvement comes from the folded cascode stage. Swapping KSA1381 etc for the MJE parts helps, but only incrementally.
If I disable the class H switching by just running both rails at the same voltage, they do single digit ppm thd. Most of the improvement comes from the folded cascode stage. Swapping KSA1381 etc for the MJE parts helps, but only incrementally.
Q9 and Q10 are the current source/sink for the second stage. They don't need speed so the MJE parts are fine. Check that you're not stuffing up the DC operating point by swapping them out. You want around 2.5mA through each. You can run this stage harder, but then you'll need to reduce the load resistors, and put the transistors on a heatsink.
Single digit PPM THD is pretty impressive especially with LatFETs, I will definitely have a look at that one thanks.
Regarding incremental gains by using KSA/KSC, I will take any advantage I can get (within reason) as it all adds up in the end 🙂
Based on sims, JFE2140 has 5.5dB more gain than SST404 in the circuit (31.4dB vs 25.9dB) but SST404 has more uncompensated BW.
When the IPS is compensated, the -3dB BW is roughly the same for both JFETs ~87.5KHz as the compensation is the limiting factor.
The extra gain of the JFE2140 does reduce the whole amplifier THD by a small amount.
More work is required to determine if that extra gain is detrimental in any way with a large input signal, I need to sim that but my thinking is that if it's a problem, more local degeneration can fix that without slowing down the stage too much with excessive compensation.
The surprise (for me) is that swapping out the cascode transistors from MMBTA06 to 2N5550 and then BC846B did not influence the gain but did increase the GBW and f3 by a lot.
This improvement is eroded by the required compensation parts C3 and R11 but hopefully this reduces THD for the IPS stage (need more sims).
See attached sim and table of results.
| All BJT | MMBTA06 | 2N5550 | BC846B | BC846B | |
| JFET pair | Gain dB | GBW MHz | GBW MHz | GBW MHz | f3 MHz # |
| JFE2140 | 31.4 | 53.2 | 106.3 | 160 | 10.6 |
| LS844/LSK489B * | 24.2 | 34.6 | 73.5 | 117.6 | 21.35 |
| LSK389B | 33.3 | 40.5 | 83.4 | 141.9 | 5.9 |
| SST401 | 24.9 | 37.5 | 80.1 | 129.6 | 21.83 |
| SST404 | 25.9 | 52.5 | 124.8 | 201.3 | 14.3 |
# f3 shown only when BC846B are used as the cascode.
I need to look at this later, but the folded cascode is one of my favorite topologies and is featured in some of the lowest distortion high speed op amps ever made - take a look at OPA891 to see what's possible 🙂
I am attaching a simulation of the amp using the TI FET and the BC846B. I will note that the models for the 2SC3503 and the 2SA1831 make a huge difference in the distortion performance. The data sheet for these parts show that they are binned for different gains. I suspect that the different models available represent different gains. To be absolutely certain the simulation is using the Cordell models I embeded them in the simulation and added a "BC" suffix (Bob Cordell).
Using Bob Cordell models didn't change the distortion much compared to using the newest models supplied by the manufacturer for me but I did have to adjust the bias of the output stage and the CC sink and sources (Q9, Q10) to set the correct current and this almost matches the THD when using BC models (although using BC models sims with less THD).
The 2/KSC and 2/KSA are indeed binned for different hFE ranges but unfortunately NPN and PNP can only be found in unmatched hFE grades for purchase.
I don't think this is too much of a problem as in every stage these transistors are used, there is a decent amount of emitter degeneration via resistors which puts a ceiling on the gain asked of the device.
Ages ago a few DIYA members measured the hFE of these parts here https://www.diyaudio.com/community/threads/2sc3503-2sa1381-e-or-f-source.237512/
The 2/KSC and 2/KSA are indeed binned for different hFE ranges but unfortunately NPN and PNP can only be found in unmatched hFE grades for purchase.
I don't think this is too much of a problem as in every stage these transistors are used, there is a decent amount of emitter degeneration via resistors which puts a ceiling on the gain asked of the device.
Ages ago a few DIYA members measured the hFE of these parts here https://www.diyaudio.com/community/threads/2sc3503-2sa1381-e-or-f-source.237512/
I've read hear on the forum somewhere that the original PCB layout for the AME6000 wasn't that great.
Could somenone explain what it is that isn't okay with the original PCB layout. Does it effect sound quality?
Has someone made an update on the PCB but still use the Hitachi's K176 and J56?
Could somenone explain what it is that isn't okay with the original PCB layout. Does it effect sound quality?
Has someone made an update on the PCB but still use the Hitachi's K176 and J56?
Interesting. I always thought the original AEM6000 was well regarded sonically, which would be inconsistent with layout issues. I have never seen anything other than a rather grainy photo of the board, so little analysis was possible. I did observe that the photo did not appear to show large bulk caps, but that would be common I think for amps of that time. Large chassis mounted caps with screw lugs were pretty much the order of the day.
With the advent of high-capacity snap mount and radial leaded caps it is more common to now see significant rail capacitance on the amplifier boards. Certainly, physics argues that when moving large currents quickly distance is your enemy, but as the complete loop for moving current to a speaker coil always includes a number of feet of speaker wire, does saving 6” between a bulk cap and an output device really improve the sound? It is not uncommon to hear different viewpoints argued passionately in these forums, sometimes with little merit.
For example, I have come to believe that blue solder resist on a PWB improves the sound quality, particularly if one primarily listens to blues and jazz. This is because blue light, being of a higher frequency, minimizes the high frequency distortion in the etch. Red solder resist on the other hand is advantageous for rock and metal. The lower frequency of red light improves the transmission of bass notes. Green solder resist? I don’t know, maybe elevator music?
With the advent of high-capacity snap mount and radial leaded caps it is more common to now see significant rail capacitance on the amplifier boards. Certainly, physics argues that when moving large currents quickly distance is your enemy, but as the complete loop for moving current to a speaker coil always includes a number of feet of speaker wire, does saving 6” between a bulk cap and an output device really improve the sound? It is not uncommon to hear different viewpoints argued passionately in these forums, sometimes with little merit.
For example, I have come to believe that blue solder resist on a PWB improves the sound quality, particularly if one primarily listens to blues and jazz. This is because blue light, being of a higher frequency, minimizes the high frequency distortion in the etch. Red solder resist on the other hand is advantageous for rock and metal. The lower frequency of red light improves the transmission of bass notes. Green solder resist? I don’t know, maybe elevator music?
@Tailgunner, I'm not aware of layout problems with the AEM6000 and as Cyberpup mentioned, It is well regarded sonically.
The sound is clean, with no audible hum or hiss with your ear right up against the tweeter.
The bass is nicer compared other amps I have used, maybe partly due to the lack of electrolytic capacitors in the feedback loop.
I also run mine without any caps on the input with my DAC so it's DC coupled via only a volume control.
It could also be because of the good damping factor.
One thing I don't like about the original layout (it wasn't a deal breaker as I still own the original modules) is how the heatsink mating surface was near the middle of the PCB and the size wasn't small so a 3-4U high rack case was required to house the modules.
I personally prefer designs where the heatsink bracket is at the edge of the PCB as this would be a lot easier to mount in a smaller case.
There was a lower power module published later that did in fact put the transistors on the edge of the PCB which made mounting easier.
SuzyJ did post the files for a TO-3 variant earlier in this thread, look at about page 7 onwards.
The sound is clean, with no audible hum or hiss with your ear right up against the tweeter.
The bass is nicer compared other amps I have used, maybe partly due to the lack of electrolytic capacitors in the feedback loop.
I also run mine without any caps on the input with my DAC so it's DC coupled via only a volume control.
It could also be because of the good damping factor.
One thing I don't like about the original layout (it wasn't a deal breaker as I still own the original modules) is how the heatsink mating surface was near the middle of the PCB and the size wasn't small so a 3-4U high rack case was required to house the modules.
I personally prefer designs where the heatsink bracket is at the edge of the PCB as this would be a lot easier to mount in a smaller case.
There was a lower power module published later that did in fact put the transistors on the edge of the PCB which made mounting easier.
SuzyJ did post the files for a TO-3 variant earlier in this thread, look at about page 7 onwards.
For those interested, I have scans of the original AEM6000 4-part article by David Tilbrook. Not sure if they were posted already.
AEM6000 4-part article by David Tilbrook.
And the status monitor, that was also a part of this project.
Thank you for posting the original schematics. There are some subtle component values differences. Also, it does have 300uF of bulk cap on each rail!
Did you make a new PCB layout for a 100W version with the Hitachi TO-3? 2SK176/J56?
I believe the version in this thread has been tuned to run cool and draw less power and uses much nicer BJTs (A1381/C3503) than the MJE 340 and 350.
Of course a different JFET and different Lateral MOSFETs have been used.
The VAS stage is also missing a pair of transistors that were removed as it's no longer differential, putting these transistors back in yields a 4-5dB reduction in H2 @ 20Khz in sims although this may not be the case in real life, at 1kHz there's probably no difference.