I'm building an integrated power amplifier. This will be based on the Trimodal power amplifier and the Precision pre-amplifier by Douglas Self. I really liked the books Douglas wrote and I decided to build this integrated amplifier. It will use to power amplifier modules based on the trimodal amplifier with four output transistors (2xNPN + 2xPNP), the updated Precision pre-amplifier with 5K linear pots and LM4562 or LME49720 opamps. I will also build the phono pre-amp with teh same opamps. The power supply will be a combination of unregulated (for the power modules) and regulated with LM317/337 for the pre-amp. So far I managed to solder the power amplifiers and I performed some measurements in class A and class B (D. Self class B 🙂 ). The results can be seen in the images below. The THD at 1kHz and 8OHM load looks very good. This is close to the limit of the Quant Asylum QA401 analyzer. Due to the size of teh heatsinks, the power amplifier will have a quiscent current of 1A. The original design by D. Self has a quiescent current of 1.6A. With 1A it can deliver around 12.5W in 8OHM. This is quite enough for my needs. In class B the quiescent current is around 50mA and the THD is also very good in my opinion. It does have a significant 2nd order distortion that I will have to look into. But for now this will have to do. I'm focusing on the other boards. I will add updates regarding the build as I go.
So I managed to assemble the power supply and measure the power amplifier powered from it. The power supply uses 2x10.000uF caps for each channel. Looking at the spectrum, I might want to add more capacitance or have a look at the design to see if I can increase the PSRR. The design itself has a very good PSRR but I guess this is what you get when using unregulated power supplies. For class A operation (first image) it definitely needs more capacitors on the power supply. I also finished the Pre-amp and will post some measurements shortly. The conclusion is that the power supply generated noise shrinks the SINAD from a respectable 91dB (class A) to 74dB. So this comes to confirm once again the importance of the power supply. In the case or unregulated power supply I guess that you can only increase the capacitance on each rail. I will experiment a bit to see what effect this has. Using regulated power supplies for a class A amplifier doe snot seem like a good approach. Another option would be to use a voltage regulator to power the input stage of the amp. In the book, D. Self finds out that most of the power supply noise comes into the amplifier via the negative rail. That is the reason why the input stage has am RC filter on the negative rail. Might try to modify this and add an LDO to see if this bring an improvement or not.
Last edited:
I finalized the preamplifier. This is based on the Precision Preamplifier and the results are shown below. I think this is very good performance and I'm really pleased with the measurements. I can't wait to actually listen to it at some point 🙂
So this is the 1KHz spectrum. This is with lowish gain and using a regulated lab power supply, not the actual one that will be used in the final build. This was measured without the tone defeat or source direct being active. The tone defeat will effectively take out the LF and HF filters, hence reducing the noise floor a bit. In any case, this is limited by the QA401 I'm using. So very pleased with it.
Next we have the same as above but with the actual regulated (LM317/LM337) power supply that will be used in the actual build. The gain was set to MAX (9Vrms). There are some mains components present (50Hz and its harmonics) but nothing audible at -100dB. That being said, I will probably have to look a bit deeper into the power supply to see if I can somehow get rid of these components or reduce them. Still very good.
Next I did a frequency response to see how the tone adjustments work. First we see the frequency response with the tone defeat active. Pretty flat. Good!
The next image is the frequency response with the tone controls set to MAX (+10dB). We can also adjust the cutoff frequency for this preamp,
So far I'm pretty please of how the different modules measure. There is still the mains noise part that I will have to investigate but other than that, it looks really good. I can't wait to actually hear how the thing sounds 🙂. Next I will measure the PHONO preamp.
So this is the 1KHz spectrum. This is with lowish gain and using a regulated lab power supply, not the actual one that will be used in the final build. This was measured without the tone defeat or source direct being active. The tone defeat will effectively take out the LF and HF filters, hence reducing the noise floor a bit. In any case, this is limited by the QA401 I'm using. So very pleased with it.
Next we have the same as above but with the actual regulated (LM317/LM337) power supply that will be used in the actual build. The gain was set to MAX (9Vrms). There are some mains components present (50Hz and its harmonics) but nothing audible at -100dB. That being said, I will probably have to look a bit deeper into the power supply to see if I can somehow get rid of these components or reduce them. Still very good.
Next I did a frequency response to see how the tone adjustments work. First we see the frequency response with the tone defeat active. Pretty flat. Good!
The next image is the frequency response with the tone controls set to MAX (+10dB). We can also adjust the cutoff frequency for this preamp,
So far I'm pretty please of how the different modules measure. There is still the mains noise part that I will have to investigate but other than that, it looks really good. I can't wait to actually hear how the thing sounds 🙂. Next I will measure the PHONO preamp.
I finished the PHONO preamp as well. Below you can see the spectrum with a 1kHz tone and the frequency response. I'm pretty pleased with the results. I guess this is as good as it gets with these op-amps (LM4562/LME49720).
IMHO, the weakest point is the cap bank. One thing I have tried with success is if you have sufficient rail voltage, a pi-filter between the output and VAS/IPS stages. In my modified amp, it originally had a 47 Ohm resistor with 47u on the VAS side. I replaced it with a pair of 22's and a two 100uF caps. It reduced the PS harmonics by 8 dB. In my amp ( 60W) I went from the original design of 3200 @ 54V on each rail to 27,000 on each rail and it is both quieter, as in can't hear a thing by the woofer or even feel the cone as well as a perceived improvement in mid-bass transients. The former can be measured, the later I think so, but do not have the equipment. I have a suspicion it would show up in a series of tone bursts.
Do remember to calculate the rectifier inrush if you increase the caps.
I have a theory, the preference for big amps is mostly for the larger power supply, not the voltage swing. A 12W amp with a big bank may sound "bigger" than a 120 W with a pathetic bank. Think in coulombs, not in uF. So lower voltage rails actually need more capacitance than high voltage rails to provide the same transient to the load. Somebody with time and equipment should test this theory.
PS, that phono preamp is incredible. It should embarrass some mega-buck stuff. Don't forget to make it easy to select input loading resistors to match your cartridge.
Do remember to calculate the rectifier inrush if you increase the caps.
I have a theory, the preference for big amps is mostly for the larger power supply, not the voltage swing. A 12W amp with a big bank may sound "bigger" than a 120 W with a pathetic bank. Think in coulombs, not in uF. So lower voltage rails actually need more capacitance than high voltage rails to provide the same transient to the load. Somebody with time and equipment should test this theory.
PS, that phono preamp is incredible. It should embarrass some mega-buck stuff. Don't forget to make it easy to select input loading resistors to match your cartridge.
Thanks for the input. Indeed the negative supply rail is where the ripple is getting in. The current design has an RC filter just before the VAS (1000uF and 15OHM). Increasing the value of the capacitor did not yield a big improvement. I will consider using an LC filter instead. I'm also considering using a small LDO to power the VAS and input stage on the negative rail. The rails are around +/-24Vdc and each channel has 2x 10.000uF. As it is now, I can't hear any hum either. Unless I really stick my era to the speaker, then I can hear a slight hum when in class A. When I will get back to the power amp I will look at improving the THD in class B as I do not like the 2nd order harmonic content. It shouldn't be there or it should be a lot smaller. I will also look at improving the PSRR but I'm not getting my hopes too high as it is already at around -100dB and I guess this is the price we pay for unregulated power supplies. I'm also testing a switched mode power supply (LLC). I'm fairly positive that the SPMPS will greatly improve the 50Hz (and its harmonics) situation. Of course I will have the ripple at the switching frequency but this is around 90-100kHz and it is a lot easier to filter. And nobody can actually hear it 🙂.
Useful inductors get insanely big. Try a pi-filter or a cap multiplier, or as you suggest, an LDO regulator. 7 ohm, cap, 7 ohm, cap, VAS
If only on one rail there may be another problem, look at your routing and dress. LM317's are dinosaurs. Newer ones from LT are many times better.
If only on one rail there may be another problem, look at your routing and dress. LM317's are dinosaurs. Newer ones from LT are many times better.
Well, usually pi filters are made out of C - L - C. What you mention is actually a 2nd order low pass filter. This will work for sure but an LC filter will yield better attenuation. Indeed for it to be relevant at 50Hz you need quite large inductors but you can also use smaller ones and target only higher frequency attenuation. There are many mains harmonics at higher frequencies and these will be thoroughly attenuated. I do agree that LM317's are dinosaurs but these are still relevant. These offer decent noise figures and PSRR. By using some extra components you can get quite good performance with a very small price. I use a 2.2R and 2x 2200uF to filter the DC coming from the bridge rectifier and I also use a 47uF capacitor on the ADJ pin to ground. Looking at ICs from LT (Analog) you can clearly find parts that are better but the price is 5-6 times higher for the LT parts. There are the TPS7A47 / TPS7A33 that have excellent performance but good luck finding one in stock 🙂. The Chinese Hi-Fi manufacturers bought the entire stock for their DACs and headphone amps 🙂. And the packaged that these come in is not that DIY friendly. I have no issues soldering these as I have access to good equipment but I can't get my hands on them 🙂. Another option is LM350 and this comes in the TO220 package and the price is close to the LM317 and there are stocks. The differences are not that big compared to LM317. There is no negative version but you can work around this by using two positive ones connected in series as long as you have separate transformer windings.
Look at the size of an inductor needed. It can be as big as the main transformer.
You already have the big C in the power supply. So basically C-R-C-R A pi filter does not have to have an inductor. Name is given by the way it shows up on a drawing.
LT regulators expensive? A 1033 ( 3A) costs about $6. Smaller ones, as we are talking the IPS and VAS, are cheaper. Price a choke large enough recently? A 1 H Hammond is $86. An independent power supply is actually cheaper.
The reason you do not usually see a regulator in this line is due to rail collapse. A passive filter will be reliable with only a volt or so drop following the rails. A regulator can drop out of regulation, or shut down with a dip in the rails.
You already have the big C in the power supply. So basically C-R-C-R A pi filter does not have to have an inductor. Name is given by the way it shows up on a drawing.
LT regulators expensive? A 1033 ( 3A) costs about $6. Smaller ones, as we are talking the IPS and VAS, are cheaper. Price a choke large enough recently? A 1 H Hammond is $86. An independent power supply is actually cheaper.
The reason you do not usually see a regulator in this line is due to rail collapse. A passive filter will be reliable with only a volt or so drop following the rails. A regulator can drop out of regulation, or shut down with a dip in the rails.
I understand what you are saying and will try it out. As for the LT1033, yes, I know this IC but it has identical performance with the LM337 (noise and ripple rejection). So it should not bring any improvement.
I now stared looking into this project as I'm done (for now) with the KT88 amp 🙂
There are two main items I wanted to tackle:
For the PSRR there are several areas where I tried to improve the circuit besides the obvious one of increasing the capacitance of the power supply output filter. I'm using now 2x10000uF. After a few attempts that did not yield any improvements, I tried using a SMPS that I bought a while ago for this precise task (https://www.audiophonics.fr/en/smps...reh-power-supply-module-300w-30v-p-14063.html). This is a +/- 30V - 300W resonant power supply. This got rid of the 50Hz (and its harmonics) as expected (see the images below).
Now moving on to the 2nd order harmonics, I got a big surprise. As I was trying all sorts of things to try and identify where the imbalance in the circuit was (as far as I know, 2nd order harmonics are generated when the amplifier is not symmetrical. Let's say that the positive cycle of the sine is a bit different from the negative one or vice-versa) I noticed that I had a lot of variations from just moving things around. I initially had the 2nd harmonic at -70dB and after moving the wires and the equipment around I had -80dB. I then started moving around the input cable (coming from the output of the QA401 and going to the input of the power amplifier). This is where I got a huge surprise. There were positions where the 2nd harmonic got below -100dB. I tried using another coax wire for the input of the amp but I got the same behavior.
I don't know what to believe now. I guess that the amp is performing this good and that there is an issue with my setup. I have no other explanation for this. Bellow you can see the THD in class AB for 1W and 5W.
Class AB - 1W into 8OHM
Class AB - 5W into 8OHM
This is the measurement before moving the input cable around:
Now, with a THD+N (SINAD) of around -92dB, this lands very close to Amir's "excellent" category. This measurement is probably limited by the QA401 as well. In class A I get the same result of -92dB, although I can see that the harmonic content is a bit better.
Class A - 5W - 8OHM load
As I'm a bit confused about the results, I'm looking at measuring using other equipment to see if this behavior is the same or not. I'll see if I can find a Cosmos E1DA around or I'll look to buy one. If you guys have any suggestions for why this THD measurement varies so much with the position of the cables, I'll be glad to hear them. I'm now doubting the THD measurement I did on the tube amp I just built 🙂
There are two main items I wanted to tackle:
- PSRR of the power amplifier
- The high 2nd order harmonics in class AB
For the PSRR there are several areas where I tried to improve the circuit besides the obvious one of increasing the capacitance of the power supply output filter. I'm using now 2x10000uF. After a few attempts that did not yield any improvements, I tried using a SMPS that I bought a while ago for this precise task (https://www.audiophonics.fr/en/smps...reh-power-supply-module-300w-30v-p-14063.html). This is a +/- 30V - 300W resonant power supply. This got rid of the 50Hz (and its harmonics) as expected (see the images below).
Now moving on to the 2nd order harmonics, I got a big surprise. As I was trying all sorts of things to try and identify where the imbalance in the circuit was (as far as I know, 2nd order harmonics are generated when the amplifier is not symmetrical. Let's say that the positive cycle of the sine is a bit different from the negative one or vice-versa) I noticed that I had a lot of variations from just moving things around. I initially had the 2nd harmonic at -70dB and after moving the wires and the equipment around I had -80dB. I then started moving around the input cable (coming from the output of the QA401 and going to the input of the power amplifier). This is where I got a huge surprise. There were positions where the 2nd harmonic got below -100dB. I tried using another coax wire for the input of the amp but I got the same behavior.
I don't know what to believe now. I guess that the amp is performing this good and that there is an issue with my setup. I have no other explanation for this. Bellow you can see the THD in class AB for 1W and 5W.
Class AB - 1W into 8OHM
Class AB - 5W into 8OHM
This is the measurement before moving the input cable around:
Now, with a THD+N (SINAD) of around -92dB, this lands very close to Amir's "excellent" category. This measurement is probably limited by the QA401 as well. In class A I get the same result of -92dB, although I can see that the harmonic content is a bit better.
Class A - 5W - 8OHM load
As I'm a bit confused about the results, I'm looking at measuring using other equipment to see if this behavior is the same or not. I'll see if I can find a Cosmos E1DA around or I'll look to buy one. If you guys have any suggestions for why this THD measurement varies so much with the position of the cables, I'll be glad to hear them. I'm now doubting the THD measurement I did on the tube amp I just built 🙂
Inductive coupling of the internal currents of the amplifier back into the input. This includes the + and - (feedback) inputs. When those currents are nonlinear, you get distortion. As far as the amplifier is concerned, these signals are indistinguishable from the desired input, and it will faithfully amplify them. Class AB generates these half-sine currents in the power supply wiring. Keep them away, and keep them at right angles to the inputs, if possible. Right angles and closer will often trump long parallel runs 6” away. It’s just as bad as hum from a transformer - same mechanism but much more insidious because you don’t realize what’s going on and know what action to take.
One of many reasons you just can’t blindly trust a simulation. And why implementation matters as much as circuit topology when you get to a certain level. When you’re trying to get -100 dB levels you have to be careful with everything.
One of many reasons you just can’t blindly trust a simulation. And why implementation matters as much as circuit topology when you get to a certain level. When you’re trying to get -100 dB levels you have to be careful with everything.
Mystery solved! I found the issue I had with the measurement setup. I overlooked one very important aspect of the QA401 analyzer, that all BNC connectors share the same ground and are not isolated from each other. The inputs and output are isolated from the enclosure and from the USB ground but not from each other. This is the same as for oscilloscopes but somehow this slipped my attention and I had ground connected on the amplifier PCB for the input cable and as well for the output one. So I had a ground loop that severely degraded the THD and THD+N measurements. What I did, I removed the ground of the coax cable connecting the output of the amplifier to the input of the QA401 analyzer. In this way the GND connection is done only through the input cable. The results can be seen below.
Class AB - 1W
Class AB - 5W
And another measurement for class AB - 5W and 96kHz bandwidth to see if the switch mode power supply introduces any nastiness at higher frequencies.
The conclusion is that the amplifier itself has excellent performance in class AB and there is very little (if any) difference when it runs in class A. Secondly, it is pretty clear that using a switching power supply is more than adequate. It will significantly improve the THD+N figure due to the lack of 100Hz (and its harmonics). The THD+N is now very close to Amir's blue category and it will probably land in it with a better analyzer. Maybe it's time to buy an E1DA Cosmos analyzer....hmmmm. I still have a few ideas on how to improve the THD even more but I will need a better method of measuring the THD if I want to improve it.
Another conclusion is that I have to redo all the measurements I did in the past (including the ones for the tube amp) with the correct measurement setup 🙂.
Class AB - 1W
Class AB - 5W
And another measurement for class AB - 5W and 96kHz bandwidth to see if the switch mode power supply introduces any nastiness at higher frequencies.
The conclusion is that the amplifier itself has excellent performance in class AB and there is very little (if any) difference when it runs in class A. Secondly, it is pretty clear that using a switching power supply is more than adequate. It will significantly improve the THD+N figure due to the lack of 100Hz (and its harmonics). The THD+N is now very close to Amir's blue category and it will probably land in it with a better analyzer. Maybe it's time to buy an E1DA Cosmos analyzer....hmmmm. I still have a few ideas on how to improve the THD even more but I will need a better method of measuring the THD if I want to improve it.
Another conclusion is that I have to redo all the measurements I did in the past (including the ones for the tube amp) with the correct measurement setup 🙂.
I have cheater plugs on the ends of my scope and sig-gen and use them as needed. The isolation trafo is reserved for the DUT (Or device with my hands inside it) - it can’t be in several places at once. I pick up a bunch of nasty 50KHz garbage from the fluorescent lights any time there’s a ground loop in my shop. It causes audible artifacts - and not just ones you need golden ears or the placebo effect to hear.
It‘s either that or turn off the lights.
It‘s either that or turn off the lights.