I am wondering where it is better to filter high frequencies. In the negative feedback loop or at the input. If one way is better than the other, then why?
In this case, I'm speaking of HF oscillations caused by connecting an AM radio to the amplifier and lots of noise is produced if the radio is near the amp or the volume is turned up.
In this case, I'm speaking of HF oscillations caused by connecting an AM radio to the amplifier and lots of noise is produced if the radio is near the amp or the volume is turned up.
Filter the input.
I suggest you double filter the input.
The main RF filter consisting of an R plus a C on the amp PCB giving an RC time constant between 300us and 1500us.
Fit an auxiliary filter right at the input socket. Try from 22pF to 100pF direct from socket barrel to socket signal. The resistance and the inductance of the interconnect and the Source create a 2pole RF filter that should not interfere with the audio signal.
The main filter should have as high an RC as you can manage without hearing any Audio Frequency attenuation (I use ~700us).
I suggest you double filter the input.
The main RF filter consisting of an R plus a C on the amp PCB giving an RC time constant between 300us and 1500us.
Fit an auxiliary filter right at the input socket. Try from 22pF to 100pF direct from socket barrel to socket signal. The resistance and the inductance of the interconnect and the Source create a 2pole RF filter that should not interfere with the audio signal.
The main filter should have as high an RC as you can manage without hearing any Audio Frequency attenuation (I use ~700us).
Hi,
You need to filter input and output, not the feedback loop.
The input stage of most chipamp's is bipolar, so it will demodulate Radio.
The ingress may come from the audio input, however I would consider speaker cables a more likely culprit (they have the right length to be good aerials for AM Radio).
If there is a capacitor across the feedback resistor, this RF will from the speaker cables will ride straight through to the differential input stage.
Sounds like your Amplifier is oscillating. Best clip on a fast 'scope and have a look.
Ciao T
You need to filter input and output, not the feedback loop.
The input stage of most chipamp's is bipolar, so it will demodulate Radio.
The ingress may come from the audio input, however I would consider speaker cables a more likely culprit (they have the right length to be good aerials for AM Radio).
If there is a capacitor across the feedback resistor, this RF will from the speaker cables will ride straight through to the differential input stage.
Sounds like your Amplifier is oscillating. Best clip on a fast 'scope and have a look.
Ciao T
The amp in question isn't even constructed yet, but my other amp has star grounding and .22 film supply decouplers as close the chip as possible. Speaker leads and input leads are kept separate and are twisted. It does not oscillate on its own (checked with scope).
OTOH, Connecting my Walkman set to AM and volume turned high enough when near the amp will cause oscillation. That amp has no RF filter. I wanted to correct that issue in the new project. I notice that some manufacturers show an RC high frequency cutoff in the neg. feedback loop (see TDA2003, 2030A and others). This is why I asked the question as input filtering seems better.
OTOH, Connecting my Walkman set to AM and volume turned high enough when near the amp will cause oscillation. That amp has no RF filter. I wanted to correct that issue in the new project. I notice that some manufacturers show an RC high frequency cutoff in the neg. feedback loop (see TDA2003, 2030A and others). This is why I asked the question as input filtering seems better.
This small cap in the NFB is part of the stability compensating system. It is used to adjust the stability margins, but misused can make the margins worse.
johnr66,
There are some ideas in Chapter 7 of ADI - Analog Dialogue | Op Amp Applications Handbook .
Cheers,
Tom
There are some ideas in Chapter 7 of ADI - Analog Dialogue | Op Amp Applications Handbook .
Cheers,
Tom
Figures 7-111 and 7-112, specifically, with 7-112 being an X2Y filter. There are caps made specifically for that---the datasheet figures for the C0G parts in Johanson's X2Y series and Freescale's app notes on the IP provide a good introduction to the filter behavior. I think 7-111 has a typo, though; R1 and R2 on the left side should be 0.5R.
In practice I've found dominant pole compensation is most interesting as a lowpass option primarily in balanced designs. In audio those usually aren't implemented with instrumentation amplifiers so a standard X2Y makes them unstable. If you want any significant filtering for RF that means pulling the feedback pole down to 50kHzish. That's not uncommon for line outputs---if you have a look at TI's differential filter app notes and DAC data sheets/eval board schematics you'll see second order MFB lowpasses cornered just above the audio range are typical of better designed differential DAC outputs---and amp theory implies it should it should be fine for power amps as well.
However, a caution. I've anecdotally found audible differences between op amps with GBPs below 50MHz whereas with faster parts I can neither hear nor measure a difference. I've not seen any rigorous study on how moving the pole affects output sound so GPB may not actually be the issue---most chip amps are pretty slow anyway (the 3886 is one of the quicker ones at 8MHz GBP typ). But I'd suggest hedging your bets by laying out a board supporting different filter options for eval rather than comitting to any one filter type.
In practice I've found dominant pole compensation is most interesting as a lowpass option primarily in balanced designs. In audio those usually aren't implemented with instrumentation amplifiers so a standard X2Y makes them unstable. If you want any significant filtering for RF that means pulling the feedback pole down to 50kHzish. That's not uncommon for line outputs---if you have a look at TI's differential filter app notes and DAC data sheets/eval board schematics you'll see second order MFB lowpasses cornered just above the audio range are typical of better designed differential DAC outputs---and amp theory implies it should it should be fine for power amps as well.
However, a caution. I've anecdotally found audible differences between op amps with GBPs below 50MHz whereas with faster parts I can neither hear nor measure a difference. I've not seen any rigorous study on how moving the pole affects output sound so GPB may not actually be the issue---most chip amps are pretty slow anyway (the 3886 is one of the quicker ones at 8MHz GBP typ). But I'd suggest hedging your bets by laying out a board supporting different filter options for eval rather than comitting to any one filter type.
in fig7-111, does the filtering C see both the resistors as parallel loads? i.e. 2r//2r = r and thus the formula for turn over frequency holds true F-3dB = 1/[2PiRC]
I always assumed I could use 2x 0.5R with a 2X value cap between, to keep the total series R the same, too. But I just simulated it, to be sure, and I was wrong! It would have to be a 2X value cap with the SAME R on both sides, not 0.5 R!
If f = 1 / 2 Pi R C gives a lowpass cutoff (-3dB) frequency for a value R, then the circuit won't have that cutoff frequency unless we use 2 x R for each of the two resistors in the "T" RCR filter, if the filter is attached to the negative input of an inverting opamp amplifier.
Oddly, maybe, this is even the case BETWEEN the resistors, when the opamp is present in the circuit. i.e. the frequency response at the top of the capacitor still has a cutoff frequency that's twice as high as the equation's result, if one side of the T RCR filter is connected to an inverting opamp amplifier's negative input pin.
If the filter is standalone, the 1 x R works like it should, of course.
So yes, Andrew, it "must" be that the resistances are seen as effectively being in parallel. Maybe I should test that by using unequal R values and checking to see if the result is equivalent to their paralleled value.
OK, I DID THAT. Interesting. I originally had two 720 Ohm resistors (with .001 uF to gnd between them). Parallel value would be 360 Ohms. So I tried 360 Ohms and 1 Meg. Putting EITHER ONE before the C, and the other one after, the result was about the same as with the two 720 Ohm resistors.
Cheers,
Tom
Last edited:
Yeah, not winding the trafo's primary leads around ferrites or providing similar filtering tends to be asking for trouble. I suspect a lot of the mains "RF" is actually SMPS backchatter but don't have data.
Assuming the load impedance is high compared to the filter impedance.
This is not the case for the R1+C+R2 topology since the op amp servos its negative input to ground. Hence the R1+C stage sees a load of R2 and doesn't behave the way you'd expect based on the simplifications used to introduce most filters.the "bottom-side fuzzies"
A few chip amps generate RF in their output stage. Short story- it's to do with lateral PNP transistors on-chip of the (bottom) output (NPN) transistor complimentary darlington. National Semiconductor calls it the "bottom-side fuzzies", where the negative going sine wave has RF oscillations due to the output darlington oscillating.
NS mentions the Zobel helps lessen these oscillations, but I find a few chip amps are a real bear to tame down, since they are making their own RF
A few chip amps generate RF in their output stage. Short story- it's to do with lateral PNP transistors on-chip of the (bottom) output (NPN) transistor complimentary darlington. National Semiconductor calls it the "bottom-side fuzzies", where the negative going sine wave has RF oscillations due to the output darlington oscillating.
NS mentions the Zobel helps lessen these oscillations, but I find a few chip amps are a real bear to tame down, since they are making their own RF
I note that NatSemi had an application note "in progress" on this specific topic when they laid off the guys who were working on the driver chips.
A few chip amps generate RF in their output stage. Short story- it's to do with lateral PNP transistors on-chip of the (bottom) output (NPN) transistor complimentary darlington. National Semiconductor calls it the "bottom-side fuzzies", where the negative going sine wave has RF oscillations due to the output darlington oscillating.
NS mentions the Zobel helps lessen these oscillations, but I find a few chip amps are a real bear to tame down, since they are making their own RF
Which ones? And which ones are the worst ones, for that?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.