Ok, I've pretty much fixed my test setup (by cutting the gnd plane in my test load). Since it has become insensitive to the connection of the scope, groundloop should be over. Very marginal improvements though. Nicer graphs as I took the time to do some linear averaging. 
So 1W becomes 0.0033% thd+n (down from 37) and 40W becomes 0.0019% (down from 24).
So 1W becomes 0.0033% thd+n (down from 37) and 40W becomes 0.0019% (down from 24).
Attachments
How about IMD? 60 Hz + 7 kHz; 18 kHz + 19 kHz; 1 kHz + 5.5 kHz, for example?
THD+N vs frequency would be interesting as well. At least spot measurements at 100, 200, 500, 1000, 2000, 5000, 10000, 20000 Hz combined into a graph.
You have a lot of ripple in your setup as indicated by the 100 Hz (and harmonics). Any plans for improving that?
Tom
THD+N vs frequency would be interesting as well. At least spot measurements at 100, 200, 500, 1000, 2000, 5000, 10000, 20000 Hz combined into a graph.
You have a lot of ripple in your setup as indicated by the 100 Hz (and harmonics). Any plans for improving that?
Tom
I re-ran the tests. I checked first at 1khz (1W and 40W) to see if everything was still fine. No changes.
So here are attached the two-tones tests. 60+7000hz (4:1) and 1Khz+5.5Khz and 18Khz+10Khz (both 1:1). I've ran them just after running the 40W test, so -5dbfs would be 40W (they're incorrectly labelled as 0dbfs=40W, sorry). The tests were ran at 48khz sampling.
I also re-ran some thd+n measurements vs frequency, at 40W. My problem is that the 18i8 distorts a lot more at 88.2Khz (or 96khz) than at 44.1khz. At 192khz, it's completely useless. On the other hand, 44.1Khz won't get me many harmonics. So my measurements setup is clearly showing its limits here. I overlapped the ranges to give an idea though.
@44.1khz
100hz : 0.0018%
200hz : 0.0016%
500hz : 0.0017%
1khz : 0.0019%
2khz : 0.0023%
5khz : 0.0034%
@88.2khz
1khz : 0.0055%
2khz : 0.0056%
5khz : 0.0063%
10khz : 0.0070%
20khz : 0.0068%
You're right, the 100hz ripple and associated noise is higher under 1Khz than I'd like. The "hash" above 1khz is in line with your spectrum for the lm3886DR though. I've yet to find the cause of that.
So here are attached the two-tones tests. 60+7000hz (4:1) and 1Khz+5.5Khz and 18Khz+10Khz (both 1:1). I've ran them just after running the 40W test, so -5dbfs would be 40W (they're incorrectly labelled as 0dbfs=40W, sorry). The tests were ran at 48khz sampling.
I also re-ran some thd+n measurements vs frequency, at 40W. My problem is that the 18i8 distorts a lot more at 88.2Khz (or 96khz) than at 44.1khz. At 192khz, it's completely useless. On the other hand, 44.1Khz won't get me many harmonics. So my measurements setup is clearly showing its limits here. I overlapped the ranges to give an idea though.
@44.1khz
100hz : 0.0018%
200hz : 0.0016%
500hz : 0.0017%
1khz : 0.0019%
2khz : 0.0023%
5khz : 0.0034%
@88.2khz
1khz : 0.0055%
2khz : 0.0056%
5khz : 0.0063%
10khz : 0.0070%
20khz : 0.0068%
You're right, the 100hz ripple and associated noise is higher under 1Khz than I'd like. The "hash" above 1khz is in line with your spectrum for the lm3886DR though. I've yet to find the cause of that.
Attachments
I hooked it up to the desktop and the juli@ rather than the 18i8. The juli@ behaves better at higher sample rates than the 18i8.
At 44.1khz, no changes at all to what was posted before.
@88.2khz (at 40W, thd+N)
1khz : 0.0036%
2khz : 0.0039%
5khz : 0.0045%
10khz : 0.0047% (missing harmonics start to show in the optimistic measurements)
20khz : 0.0034% (once again, unrealistic due to missing harmonics)
@192khz, thd+n is stuck at 0.017%, whatever the frequency. If I just take thd, it goes like this:
1khz: 0.0018% (as a reference, thd at 1Khz with 44.1khz sampling is 0.0013%)
2khz: 0.0020%
5khz: 0.0030%
10khz: 0.0042%
20khz: 0.0038% (once again, lack of proper bandwidth starts to show)
At 44.1khz, no changes at all to what was posted before.
@88.2khz (at 40W, thd+N)
1khz : 0.0036%
2khz : 0.0039%
5khz : 0.0045%
10khz : 0.0047% (missing harmonics start to show in the optimistic measurements)
20khz : 0.0034% (once again, unrealistic due to missing harmonics)
@192khz, thd+n is stuck at 0.017%, whatever the frequency. If I just take thd, it goes like this:
1khz: 0.0018% (as a reference, thd at 1Khz with 44.1khz sampling is 0.0013%)
2khz: 0.0020%
5khz: 0.0030%
10khz: 0.0042%
20khz: 0.0038% (once again, lack of proper bandwidth starts to show)
I'm still searching the source of my 100hz ripple. Without much success I've got to say.
I first tried bypassing the protection board. No difference. Btw, it showed that the protection board doesn't contribute to distortion (within the limits of my setup).
The, I tried shorting out the 10R resistor I've got between signal gnd and output gnd. No difference wrt ripple. Interestingly no improvement to thd+N but sadly none to ripple either.
I also checked the ripple on the rails. At no load, 50mV. At higher load (one channel, 40W/8R/1khz), 0.5V.
Moving wires around doesn't help none. I'm running out of ideas. Maybe the big input caps could pick up noise ? Or the 90° spades I'm using ?
I first tried bypassing the protection board. No difference. Btw, it showed that the protection board doesn't contribute to distortion (within the limits of my setup).
The, I tried shorting out the 10R resistor I've got between signal gnd and output gnd. No difference wrt ripple. Interestingly no improvement to thd+N but sadly none to ripple either.
I also checked the ripple on the rails. At no load, 50mV. At higher load (one channel, 40W/8R/1khz), 0.5V.
Moving wires around doesn't help none. I'm running out of ideas. Maybe the big input caps could pick up noise ? Or the 90° spades I'm using ?
Did you try regulated supply?
Or you can add LC filter after your PSU C bank, you can get toroid type 26 cores cheap or from broken PC PSU-u and make your own L.
@Mark: that's my guess too. Whatever I'm doing in the amplifier has very little impact on the 100hz showing up in Arta (at about -105dB under the signal, whatever the power output). Having the PC on mains or battery doesn't change anything either. The 18i8 is mains powered, with a two prongs adapter.
I attach the schematic of the test setup. The dc-dc converter is isolated btw. In loopback there isn't any problem. There are two jumpers to tie the shield from the balanced cables coming from/going to the soundcard. If it let them both open, 50hz pick up increase a bit but nothing changes for 100hz. Having either a 1k resistor or a 0R jumper at the jumper for the cable going to the soundcard input fixes the 50hz.
I also attach a diagram of the amp setup, with only one channel wired up. Shorting the HBR doesn't do anything nor does removing the link from the power ground to chassis, nor does moving the speaker return to the PS.
Finally, I attach the layout of the PS and amp.
Any suggestions on what to try ? I don't have a regulated bench supply at hand right now.
I attach the schematic of the test setup. The dc-dc converter is isolated btw. In loopback there isn't any problem. There are two jumpers to tie the shield from the balanced cables coming from/going to the soundcard. If it let them both open, 50hz pick up increase a bit but nothing changes for 100hz. Having either a 1k resistor or a 0R jumper at the jumper for the cable going to the soundcard input fixes the 50hz.
I also attach a diagram of the amp setup, with only one channel wired up. Shorting the HBR doesn't do anything nor does removing the link from the power ground to chassis, nor does moving the speaker return to the PS.
Finally, I attach the layout of the PS and amp.
Any suggestions on what to try ? I don't have a regulated bench supply at hand right now.
Attachments
I tried the two 9V batteries, no change.
I tried another amp, exactly the same level of 100hz (105dB down).
I tried unplugging. It stops real fast (muting ?) but during the split second I've got a signal, the 100hz component seems to drop quite a bit.
I'm considering adding a 100r resistor in one of those two positions...
Edit: I also tried an audio transformer at the input of the amp. It picks up 50hz like crazy and thd is badly worsened... but it has 0 impact on the 100hz ripple.
I tried another amp, exactly the same level of 100hz (105dB down).
I tried unplugging. It stops real fast (muting ?) but during the split second I've got a signal, the 100hz component seems to drop quite a bit.
I'm considering adding a 100r resistor in one of those two positions...
Edit: I also tried an audio transformer at the input of the amp. It picks up 50hz like crazy and thd is badly worsened... but it has 0 impact on the 100hz ripple.
Attachments
Last edited:
Hi 00940, first thank you for all the effort you have put into this project. The fact that you trying to help others without making a buck yourself is refreshing. The amp boards look good and I have already placed my order for 10 @ $2. Any concern you had with JLCPCB is unfounded, I have used them before and their quality is awesome in my opinion.I am surprised and a little dissappointed that more people havent given your project a try. I am interested in further feedback on the final result. Keep up the good work!!
Thx for the good words.
There a few good reasons I can think of why it has become slow here:
- I've yet to really post a clear BOM, clear documentation, etc. It will not come before end of August as I'm quite busy these days (leading a group of pilgrims in Spain, I've only my phone with me).
- I promised to redo tests with the Millet interface to see if the 100hz artifacts really came from my testing setup. That will have to wait also until end of August.
- I also promised to redo the PS and DC protection boards that needed a quick touch up. I still have to post that too.
So I'm not disappointed in anyone but myself at this point.
There a few good reasons I can think of why it has become slow here:
- I've yet to really post a clear BOM, clear documentation, etc. It will not come before end of August as I'm quite busy these days (leading a group of pilgrims in Spain, I've only my phone with me).
- I promised to redo tests with the Millet interface to see if the 100hz artifacts really came from my testing setup. That will have to wait also until end of August.
- I also promised to redo the PS and DC protection boards that needed a quick touch up. I still have to post that too.
So I'm not disappointed in anyone but myself at this point.
Hi 00940,
Like Quadfan, I'm also appreciative of this project. As you rightly note, people may be waiting for a 'final' iteration with the full BOM etc, I know I'm one of those as I'm still very much a novice. It has been interesting following the discussion though - so, many thanks for initiating this.
Cheers
Neal
Like Quadfan, I'm also appreciative of this project. As you rightly note, people may be waiting for a 'final' iteration with the full BOM etc, I know I'm one of those as I'm still very much a novice. It has been interesting following the discussion though - so, many thanks for initiating this.
Cheers
Neal