I saved the graphs from the tests I ran yesterday and if I compare them side by side, the loaded IMD graph has more "spikes" compared to the unloaded one. Anyways, here are the high gain results. The volume level was about 38%, if I recall right. Most of the numbers seem lower. Noise performance is degraded, which is logical, since it was run at high gain (4.8x). First one is the unloaded test and second the loaded one. I have to say I'm happy with the results so far. They seem pretty close to the performance the X-Fi, when the amp is using unity gain.
Update to the loaded high gain RMAA test
I would have edited the previous post and added the updated RMAA results, but I can't find a edit option in the posts. I'm I just blind, or is it missing? So, sorry for the spamming.
I had scribbled down the volume setting down on a piece of paper, and it was actully 30% not 38% as I read. I reran the test and found that 40% output was the absolute maximum the input could take. I also paid more attention to the unity gain test settings (input level) and ran that also a second time. Here are the results, high gain on the left and unity on the right. The curious thing is, that the distortion numbers are lower and crosstalk is better with high gain. Is that a result of R3 being added to the circuit? The noise performance is worse, as there is gain, which seems logical to me.
I would have edited the previous post and added the updated RMAA results, but I can't find a edit option in the posts. I'm I just blind, or is it missing? So, sorry for the spamming.
I had scribbled down the volume setting down on a piece of paper, and it was actully 30% not 38% as I read. I reran the test and found that 40% output was the absolute maximum the input could take. I also paid more attention to the unity gain test settings (input level) and ran that also a second time. Here are the results, high gain on the left and unity on the right. The curious thing is, that the distortion numbers are lower and crosstalk is better with high gain. Is that a result of R3 being added to the circuit? The noise performance is worse, as there is gain, which seems logical to me.
This is peculiar, no doubt. NwAvGuy's results for a 4556A-based cMoy (unity gain, gain 4.6X) suggest nothing of the sort, with unity gain clearly giving better results than high gain.
A few theories:
1. It seems the X-Fi output has less noise but somewhat higher distortion than the input.
2. What does the crosstalk graph look like? Capacitive coupling yields levels rising at 6 dB/octave, while board contamination (e.g. flux residue) would give a response that's about constant with frequency. If the former dominates, how big is your (-in) node?
3. The difference might be related to input impedance distortion. Could you measure again in unity gain with 10k replaced by 2.6k? (Maybe do only one channel and ask RMAA to display "details", where distortion values for both channels are shown.) If that makes a difference, try matching the impedances seen by the (+in) and (-in) nodes as precisely as possible (with the pot all the way up, obviously - a match should be obtained at about -6 dB now but that would be of no use here).
4. The difference might be related to common-mode distortion. But then, why would distortion worsen under load?
5. The opamp might be on the verge of oscillation - this is always more critical at low gains. Adding some output series resistance (10-22 ohms should do) ought to make a noticeable difference then. Note that (-in) node extension and associated impedance also plays a role due to stray capacitance (you do the math on what several pF from (-in) to ground will do), hence why feedback network resistors tend to be placed as close as possible to the opamp. This is why it is better to first have the 2.6k and then the gain switch to ground rather than vice versa.
I rather doubt it's 3., since the tiny input bias current would point towards very small input transistors at low tail current, but you never know. Unfortunately the NJM4556A was one of the notable omissions when Samuel Groner took his series of opamp measurements.
Should you happen to be extremely bored for some reason, you could also try an inverting version of the circuit, which does away with several distortion mechanisms. Inverting merely swaps two nodes: Former feedback network ground becomes signal input, and former signal input becomes ground.
In inverting operation, feedback turns the inverting input (-in) into a current input - the circuit is now an I/V or transimpedance amplifier (current in, voltage out, with a "gain" that has a dimension of voltage/current = transimpedance, in ohms obviously). The input resistor does the V/I part so the whole circuit becomes a voltage amplifier again. Since (ideally) the opamp takes up no current, it is the same in both feedback and input resistor and their ratio determines gain.
Current input means it's a virtual short, so both inputs remain at the same potential, and if the (+in) is at ground, this is it. Hence there is no signal-related common-mode voltage appearing there, and input impedance distortion has no chance when both input nodes are low impedance.
The edit time limit is half an hour here. I've always found that to be a bit on the short side.
A few theories:
1. It seems the X-Fi output has less noise but somewhat higher distortion than the input.
2. What does the crosstalk graph look like? Capacitive coupling yields levels rising at 6 dB/octave, while board contamination (e.g. flux residue) would give a response that's about constant with frequency. If the former dominates, how big is your (-in) node?
3. The difference might be related to input impedance distortion. Could you measure again in unity gain with 10k replaced by 2.6k? (Maybe do only one channel and ask RMAA to display "details", where distortion values for both channels are shown.) If that makes a difference, try matching the impedances seen by the (+in) and (-in) nodes as precisely as possible (with the pot all the way up, obviously - a match should be obtained at about -6 dB now but that would be of no use here).
4. The difference might be related to common-mode distortion. But then, why would distortion worsen under load?
5. The opamp might be on the verge of oscillation - this is always more critical at low gains. Adding some output series resistance (10-22 ohms should do) ought to make a noticeable difference then. Note that (-in) node extension and associated impedance also plays a role due to stray capacitance (you do the math on what several pF from (-in) to ground will do), hence why feedback network resistors tend to be placed as close as possible to the opamp. This is why it is better to first have the 2.6k and then the gain switch to ground rather than vice versa.
I rather doubt it's 3., since the tiny input bias current would point towards very small input transistors at low tail current, but you never know. Unfortunately the NJM4556A was one of the notable omissions when Samuel Groner took his series of opamp measurements.
Should you happen to be extremely bored for some reason, you could also try an inverting version of the circuit, which does away with several distortion mechanisms. Inverting merely swaps two nodes: Former feedback network ground becomes signal input, and former signal input becomes ground.
In inverting operation, feedback turns the inverting input (-in) into a current input - the circuit is now an I/V or transimpedance amplifier (current in, voltage out, with a "gain" that has a dimension of voltage/current = transimpedance, in ohms obviously). The input resistor does the V/I part so the whole circuit becomes a voltage amplifier again. Since (ideally) the opamp takes up no current, it is the same in both feedback and input resistor and their ratio determines gain.
Current input means it's a virtual short, so both inputs remain at the same potential, and if the (+in) is at ground, this is it. Hence there is no signal-related common-mode voltage appearing there, and input impedance distortion has no chance when both input nodes are low impedance.
The edit time limit is half an hour here. I've always found that to be a bit on the short side.
Last edited:
Thank you for the input, sgrossklass! Nwavguys blog was the the text that gave me original push for the idea to build the cMoy. I was very impressed by the numbers the simple amp gave equipped with the 4556 op amp.
I'll dig into the unity gain problem, once I have a bit more time. For the time being I'll use the high gain setting, since I can make the incoming signal lower quite easily.
I uploaded all the graphs from the RMAA runs, here is unity gain data and here the high gain data. The source graphs (X-Fi) are here.
The gain switch is wired in the following manner: I have R3 (2.6k) soldered from one leg to the feedback network and the other leg is soldered to the wire leading to the gain switch. The switch has then a wire connected to ground.
I'll dig into the unity gain problem, once I have a bit more time. For the time being I'll use the high gain setting, since I can make the incoming signal lower quite easily.
I uploaded all the graphs from the RMAA runs, here is unity gain data and here the high gain data. The source graphs (X-Fi) are here.
The gain switch is wired in the following manner: I have R3 (2.6k) soldered from one leg to the feedback network and the other leg is soldered to the wire leading to the gain switch. The switch has then a wire connected to ground.
Ok, I think I got it solved now. The board was clean, I had cleaned it thoroughly when I built it. Since it seemed I would have to resort to the solder iron, I pondered my options. The series resistor would have been almost impossible to fit in nicely, so I decided the next option would adding a second set of R3, giving me a small amount of gain. I had to use another switch, but I had already one fit for the task. So I went ahead and added two more optional R3 resistors, a matched set of 33Ks (32,6K each to be precise). That gave me gain of about 1.3. I fired up the cmoy and measured the offset, left channel being below what my DMM could measure and the right giving me 0.01 mV. All good there. Then I ran RMAA, and it seems the problem went away. The numbers are pretty consistent with what the X-Fi puts out unloaded (screen shots below). Maybe the op amp was close to oscillation and a little bit of gain was all that was needed, at least it seems so.
A big thank you for sgrossklass for pointing out possible reasons to the problem.
A big thank you for sgrossklass for pointing out possible reasons to the problem.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.