I recently replaced the cheaper resistors and caps in the signal path in my LM3875 chip amp with more expensive parts: Vishay S102K for the 10k resistor to the - input, Caddock 132 (332kohm) resistor for feedback, Axon input cap. The result, near disaster. Perhaps that is overstating it, but the amplifier definitely sounded better with the cheaper components.
So, I took the amp to the scope and started looking around for trouble. Gain vs. freq.: perfect. Max. ouput signal: Fine. I was looking at high-frequency roll-off and I cranked up the frequency of the sine wave to 100 kHz. Here's where the strangeness lies. The output wave was phase-shifted from the input wave. Okay, I think, maybe that is just the phase-shift due to the GBW product.
Then I looked at a square wave and with a small-signal input the response was nearly indistinguishable from an RC low-pass filter - extremely overdamped; I hadn't installed a cap across the feedback resistor. With a large signal input, the slew rate was fine at about 15 V/micro-second, so it wasn't slew-rate limited.
Has anyone else observed this behavior?
I went back to check the phase-shift more closely. It turns out that there was about a 1.1 micro-second delay between the input and the output signals at all frequencies I measured, down to about 15kHz, where the gain is still flat. It was difficult to measure the phase delay accurately below 15 kHz. I am not sure of the Caddock MK-132 construction, but if it is like the Vishay, it has a signal path in the resistor that is longer for higher resistance values. The Vishay data sheet says 1kohm resistors have a 1 nano-second delay. So, I paralleled lower value carbon film resistors with the feedback resistor and, voila, the delay decreases to about 0.3 micro-seconds with a 44kohm in parallel with the 332 kohm resistor.
I haven't yet taken it apart to replace the feedback resistor, since I don't have the right value on hand.
Has anyone else had similar trouble with high-value Caddock or Vishay resistors? It kind of defeats the purpose of a very short feedback path length if the resistor adds 80 meters of wire from lead to lead.
Jeremy
So, I took the amp to the scope and started looking around for trouble. Gain vs. freq.: perfect. Max. ouput signal: Fine. I was looking at high-frequency roll-off and I cranked up the frequency of the sine wave to 100 kHz. Here's where the strangeness lies. The output wave was phase-shifted from the input wave. Okay, I think, maybe that is just the phase-shift due to the GBW product.
Then I looked at a square wave and with a small-signal input the response was nearly indistinguishable from an RC low-pass filter - extremely overdamped; I hadn't installed a cap across the feedback resistor. With a large signal input, the slew rate was fine at about 15 V/micro-second, so it wasn't slew-rate limited.
Has anyone else observed this behavior?
I went back to check the phase-shift more closely. It turns out that there was about a 1.1 micro-second delay between the input and the output signals at all frequencies I measured, down to about 15kHz, where the gain is still flat. It was difficult to measure the phase delay accurately below 15 kHz. I am not sure of the Caddock MK-132 construction, but if it is like the Vishay, it has a signal path in the resistor that is longer for higher resistance values. The Vishay data sheet says 1kohm resistors have a 1 nano-second delay. So, I paralleled lower value carbon film resistors with the feedback resistor and, voila, the delay decreases to about 0.3 micro-seconds with a 44kohm in parallel with the 332 kohm resistor.
I haven't yet taken it apart to replace the feedback resistor, since I don't have the right value on hand.
Has anyone else had similar trouble with high-value Caddock or Vishay resistors? It kind of defeats the purpose of a very short feedback path length if the resistor adds 80 meters of wire from lead to lead.
Jeremy
This doesn't sound abnormal to me. It seems like a normal result of the internal frequency compensation of the chip amplifier.
Even with perfect resistors, if you use a conventional operational amplifier or a chip amp at a closed-loop gain considerably higher than the minimum for which its stability is guaranteed, it is normal that the circuit has approximately a first-order behaviour, with a small-signal step response with approximately a
(1-exp(-t/tau))
shape.
Theoretically, the time constant should be equal to
tau=(R1/R2+1)/(2*pi*GBP),
where R1 is the feedback resistance from the output to the negative input, R2 is the resistance from the negative input to ground or to the signal source, GBP is the gain-bandwidth product (in Hz) and pi is 3.14159265358979...
If the amplifier is well-designed, the step response will gradually change into a well-damped second order response when you reduce the closed-loop gain to the minimum recommended value.
Even with perfect resistors, if you use a conventional operational amplifier or a chip amp at a closed-loop gain considerably higher than the minimum for which its stability is guaranteed, it is normal that the circuit has approximately a first-order behaviour, with a small-signal step response with approximately a
(1-exp(-t/tau))
shape.
Theoretically, the time constant should be equal to
tau=(R1/R2+1)/(2*pi*GBP),
where R1 is the feedback resistance from the output to the negative input, R2 is the resistance from the negative input to ground or to the signal source, GBP is the gain-bandwidth product (in Hz) and pi is 3.14159265358979...
If the amplifier is well-designed, the step response will gradually change into a well-damped second order response when you reduce the closed-loop gain to the minimum recommended value.
Thanks for the details about the delay. That makes good sense. It surprised me because I hadn't seen such an obvious shift from input to output before. Of course, I have worked mostly with DC in the past.
Assuming the National Semiconductor literature to be accurate, the numbers work out to be very close to the measured 1.1 microsecond delay observed.
That leaves the question of the poor sound... I'll have to go back and change one thing at a time and see what the effects are.
Jeremy
Assuming the National Semiconductor literature to be accurate, the numbers work out to be very close to the measured 1.1 microsecond delay observed.
That leaves the question of the poor sound... I'll have to go back and change one thing at a time and see what the effects are.
Jeremy
Hi Jeremy,
Something like 1us delay is exactly what I see running the LM1875 (similar GBW) model. This delay goes down together with the amps gain. However, I am not sure why there will be such a difference between small and large signal pulse response that you found. That is interesting.
Joe Rasmussen found strange behavior above 320kHz and that started to happen rapidly at 330kHz.
http://members.ozemail.com.au/~lisaras/design.htm
(scroll down to the bottom half of the page)
Note that he uses a higher gain (1M/(18k+4.7k+Zout) feedback). If this freq would be shifted with lower gain, I’d rather expect to see it shifted up.
If you repeat that measurement with 100kHz (not bad to go even higher) using the cheaper/other resistors, please, keep us informed about it. It will be interesting to know if anything of this is related to the used (type or brand of the) resistor.
Pedja
Something like 1us delay is exactly what I see running the LM1875 (similar GBW) model. This delay goes down together with the amps gain. However, I am not sure why there will be such a difference between small and large signal pulse response that you found. That is interesting.
Joe Rasmussen found strange behavior above 320kHz and that started to happen rapidly at 330kHz.
http://members.ozemail.com.au/~lisaras/design.htm
(scroll down to the bottom half of the page)
Note that he uses a higher gain (1M/(18k+4.7k+Zout) feedback). If this freq would be shifted with lower gain, I’d rather expect to see it shifted up.
If you repeat that measurement with 100kHz (not bad to go even higher) using the cheaper/other resistors, please, keep us informed about it. It will be interesting to know if anything of this is related to the used (type or brand of the) resistor.
Pedja
One thing to keep in mind is that Bulk Metal Foil Resistors are highly inductive while carbon composition and Metal Oxide are relatively low inductance. Bulk Metal Foil resistors while great at noise specs are not the best choice for feedback networks. This can cause time delay in the larger values because the self inductance of the resistor is larger at higher values and can get to the point of reacting with the parasitic capacitance both on Board and on the Chip to create a L/c resonant circuit.
I think, reading all the answers, that kropf really must do a comparison with a "normal" resitor. The thread is called "what a difference ....", but I cannot find what this difference is, if there is any. Without that, any cause attributed to the feedback resistor remains pure speculation, if not fantasy.
Jan Didden
Jan Didden
it appears that the caddock Bulk metal foil resistors are rather low in inductence the 10 nH inductence is real Low and on par with the Wima Polopropylene Box capacitors often used for bypassing thay also have about 10 nH of inductence the caddock resistors can be found at
http://www.caddock.com/Online_catalog/Mrktg_Lit/MP9000_Series.pdf
http://www.caddock.com/Online_catalog/Mrktg_Lit/MP9000_Series.pdf
Jan,
This thread came to life again before I had a chance to measure the circuit with "normal" resistors. Yes, I absolutely intend to make that measurement. Unfortunately it won't be easy, since the feedback resistor is nearly inaccessible and I don't want to destroy the expensive components around it.
Until then, in the absence of measurement, I do have two data points with the Caddock, which means that if I assume a time delay in the resistor, I can solve the two equations for the IC's actual GBP (1.1us delay at 330kohm, 0.3us delay at 44k||330k). The time delays I measured are accurate to better than 0.05 us. Solving the two equations, I come up with a GBP of about 5.8MHz and a delay in the resistor of 150+/-50ns. I hope this will be confirmed by actual measurement in a few days.
Jeremy
This thread came to life again before I had a chance to measure the circuit with "normal" resistors. Yes, I absolutely intend to make that measurement. Unfortunately it won't be easy, since the feedback resistor is nearly inaccessible and I don't want to destroy the expensive components around it.
Until then, in the absence of measurement, I do have two data points with the Caddock, which means that if I assume a time delay in the resistor, I can solve the two equations for the IC's actual GBP (1.1us delay at 330kohm, 0.3us delay at 44k||330k). The time delays I measured are accurate to better than 0.05 us. Solving the two equations, I come up with a GBP of about 5.8MHz and a delay in the resistor of 150+/-50ns. I hope this will be confirmed by actual measurement in a few days.
Jeremy
Jeremy,
I understand, and please note that I don't want to sound negative or anything. But suppose that you measure the same with a 2 ct regular film resistor? This is very possible, the delay you measure may be due to PCB stray capacitance. So, if you want to know what the expensive resistor does in this respect, you can't get around a comparison.
Jan Didden
I understand, and please note that I don't want to sound negative or anything. But suppose that you measure the same with a 2 ct regular film resistor? This is very possible, the delay you measure may be due to PCB stray capacitance. So, if you want to know what the expensive resistor does in this respect, you can't get around a comparison.
Jan Didden
After inadvertantly melting some plastic wire insulation with the fat part of the soldering iron, I was able to remove the Caddock resistor and replace it with an el cheapo carbon film of approximately the same value.
The result? No difference in the delay observed. Although, it was fun observing the wild variation in delay when I only had the new resistor connected using test clips. I got more than a 2 micro-second delay by winding the two test clip leads together over their 18" length.
My apologies for floating this red herring.
Jeremy
The result? No difference in the delay observed. Although, it was fun observing the wild variation in delay when I only had the new resistor connected using test clips. I got more than a 2 micro-second delay by winding the two test clip leads together over their 18" length.
My apologies for floating this red herring.
Jeremy
It kind of defeats the purpose of a very short feedback path length if the resistor adds 80 meters of wire from lead to lead.
Me thinks so. Hmm, 80 meters. Probably add a good delay to the feedback signal. I basically soldered the feedback resister to the leads of the chip. The loop is about 10mm, if that. I think I used a carbon film, so the length you see is the lenght you get. Wirewound seems counter productive here. I wonder if all that length picks up any extra noise.
Cheers,
Cyber
Me thinks so. Hmm, 80 meters. Probably add a good delay to the feedback signal. I basically soldered the feedback resister to the leads of the chip. The loop is about 10mm, if that. I think I used a carbon film, so the length you see is the lenght you get. Wirewound seems counter productive here. I wonder if all that length picks up any extra noise.
Cheers,
Cyber
What in the hell.........?
"So, I paralleled lower value carbon film resistors with the feedback resistor and, voila, the delay decreases to about 0.3 micro-seconds with a 44kohm in parallel with the 332 kohm resistor."
I think it is the absolute value (not the ratio of the feedback less there be any confusion in what I am saying ) of the resistors that is the problem.... not the parasitics in the resistors. ? The amp has input capacitances at the feedback terminal that roll off the signal more quickly with increasing values of R. Also the voltage noise will be greater (current noise times resistance).
Start with resistor values similar to those the app notes. That's why they write app notes. BTW the lead and trace inductance will most likely be greater than the parasitic indutance of the resistors. I did get a good laugh when I read this thread. I hope John Curl doesn't read it and hurt himself.....
"So, I paralleled lower value carbon film resistors with the feedback resistor and, voila, the delay decreases to about 0.3 micro-seconds with a 44kohm in parallel with the 332 kohm resistor."
I think it is the absolute value (not the ratio of the feedback less there be any confusion in what I am saying ) of the resistors that is the problem.... not the parasitics in the resistors. ? The amp has input capacitances at the feedback terminal that roll off the signal more quickly with increasing values of R. Also the voltage noise will be greater (current noise times resistance).
Start with resistor values similar to those the app notes. That's why they write app notes. BTW the lead and trace inductance will most likely be greater than the parasitic indutance of the resistors. I did get a good laugh when I read this thread. I hope John Curl doesn't read it and hurt himself.....
Your right but........
I was quite content to be very nice until I read this bit of "Rollinesque" ad libbing. I credit kropf for quickly finding his error but I still don't think there is 100nS of delay in his resistors.
BTW I have seen large value wirewounds used in tube amplifiers (Audio Research) without problems.
ppl said:
I was quite content to be very nice until I read this bit of "Rollinesque" ad libbing. I credit kropf for quickly finding his error but I still don't think there is 100nS of delay in his resistors.
BTW I have seen large value wirewounds used in tube amplifiers (Audio Research) without problems.
Re: Your right but........
Imagine my surprise that this thread came back to life. Fred, thanks for the credit. However, the resistor could have a 100 ns delay. I haven't cut one open to check, but the description and the form factor indicate that it could be similar to the Vishay S102 series bulk metal foil resistor. Vishay does provide a detailed data sheet. The resistor is made up a very long metal wire, with a induction-cancelling path (yes that does defeat the purpose of a short feedback line). For a 1k resistor, Vishay specs a 1 ns delay from the length of the wire, about 1 ns every 9" length. So, it is possible that a 330 k resistor has a 2970" wire as the resistive element - probably not, it is likely a bit thinner as well.
Jeremy
Fred Dieckmann said:
Imagine my surprise that this thread came back to life. Fred, thanks for the credit. However, the resistor could have a 100 ns delay. I haven't cut one open to check, but the description and the form factor indicate that it could be similar to the Vishay S102 series bulk metal foil resistor. Vishay does provide a detailed data sheet. The resistor is made up a very long metal wire, with a induction-cancelling path (yes that does defeat the purpose of a short feedback line). For a 1k resistor, Vishay specs a 1 ns delay from the length of the wire, about 1 ns every 9" length. So, it is possible that a 330 k resistor has a 2970" wire as the resistive element - probably not, it is likely a bit thinner as well.
Jeremy
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.