# DC blocking capacitor in opamp feedback loop - does this count as the “signal path”?

I'm a bit new to electronics, so this may be a dumb question, though I hope not.

I'm working on a pre-amp circuit which can drive several TDA7492 boards from a single source (I've found they don't enjoy all being connected directly to some sources)

The preamp is a single supply design, with a gentle gain to boost the 1.25v line level from my Pi to "professional" +4dB level (These boards can take up to 3.6v peak-peak) to maximise the signal, followed by the amps themselves and a unity-gain stage to supply an isolated aux-out for my friend's poweramp.

It works nicely in PartSim, though I haven't decided exactly on some of the component values for the gain: http://partsim.com/simulator/#38256

/Edit: Added a much simplified schematic:

The coupling caps will be polyester, and I've arranged so that I can use low values, which are within budget. Higher value polys get expensive very quickly. I've had all the expected issues with version 1.0 of the amp where I very mistakenly used Tantalums. I understand now why these are very bad - inductance and ESR.

My problem is with C1 in the circuit. What C1 does is block DC and thus sets the DC gain to unity, which keeps the DC offset at 1/2 supply whatever AC gain is applied. Without this, the gain pushes the offset up and can clip the waveform.

I need the feedback resistance to be low overall, as I've read that this is good for low distortion - I assume the distortion would be due to thermal noise in the feedback circuit. This means I need a high value cap here so that the filtering effect of R1+R2 & C1 doesn't lower the gain in the audio range. The best I can achieve is the 220uF you see here which drops the output by 10% between 20 and 50Hz. I can live with this.

But my problem is understanding whether C1 is in the signal path. I can see arguments both ways - the signal doesn't really flow through it, so probably not. But if I used an electrolytic here and it didn't pass the signal cleanly, would this in fact create distortion before the capacitor and pass the noise in to the inverting input, either affecting the gain or even amplifying the distortion?

If I can use an electrolytic, that's great. If I can't, I've got a problem with a high enough value poly cap being extremely expensive (But perhaps I can compromise here - can anyone suggest a more appropriate, inexpensive capacitor type?)

Thanks very much in advance! :-)

-Oli

• Draw a circuit. – Andy aka Nov 27 '15 at 12:45
• @Andyaka the link to the partsim contains the schematic. I'm not a fan of clicking links to get essential parts of the question though... – Arsenal Nov 27 '15 at 12:56
• The circuit is a mess - try redrawing it neatly so it can be read when you eventually copy and paste it to this question. – Andy aka Nov 27 '15 at 13:05
• Sorry chaps, wasn't aware of the etiquette... greatly simplified schematic added. In this one, it is C1 & R1 + R2 that are the potential problem. – Oli Comber Nov 27 '15 at 13:24
• If you're worried about the high ESR (Equivalent Series Resistance) in an electrolytics cap, note that it will be swamped by the massively higher 3300 ohm series resistance you have deliberately added in R2. – pipe Nov 27 '15 at 14:38

C1 does affect the signal path but your premise about keeping the feedback resistors low in value is largely unfounded. The op-amp has an equivalent input noise of 66nV per sqrt(Hz) and this may sound confusing but to help you out I'll explain. The bandwidth is audio hence 20kHz as near as damn it so take the sq root of 20,000 to get 141 and multiply this by the 66nV to get the relevant audio noise produced by the op-amp - I estimate 9.3 uV RMS.

The thermal noise of a 3300 ohm resistor at 20 degC is ~1 uV RMS but you have, in effect two in parallel as far as AC is concerned so that's a noise of 0.73 uV. Use this calculator to try this your self.

So you have about 0.73 "adding" to 9.3 and you have to add noises vectorially by pythagourous: -

Noise = $\sqrt{0.73^2 + 9.3^3}$ = 9.33 uV i.e. SFA difference. So you could choose 33k resistors and these together have a parallel resistance of 16.5k and produce a thermal noise of 2.3 uV. When pythagorus has his way the net voltage noise is 9.58 uV or 0.26 dB more noise.

If two 330ks are used the net noise is 7.3 uV and combining this with the op-amp gives a total voltage noise at the input to the op-amp of 11.82 uV. This is about 2.1 dB more noisier than the op-amp on its own.

Hopefully you can see that you can choose much bigger resistors than what you thought and use maybe a 10uF ceramic.

I'll also add that your original 220 uF capacitor and 3300 ohm resistor will give a 3dB point at about 0.22 Hz - this for audio is ridiculous and a capacitor that is ten times smaller at 22 uF is much more suited. This all leads me to say that 33k resistors and a 2.2 uF cap is all that is needed.

The TL064 is a better choice regards input noise but it aint perfecto. For instance, the TL064 has a guaranteed input common mode range of +/-11V on a +/-15V rail. This means that on a +/-6V rail (12V and 0V is the same), the input signal level has to be biased at 6V and cannot reasonably be greater in amplitude than 4Vp-p. However, the main problem is the output amplitude - the DS says it is guaranteed to produce +/-10V p-p on a +/-15V rail and, transcribing this to a 12V single rail means you are only going to be seeing a guaranteed 2Vp-p. Of course these numbers are the extremes but, if you were building many of these devices you would choose a better op-amp.

• That's fantastic, thanks very much Andy! I've changed this one to my accepted answer. That's taught me a lot I wasn't expecting about noise and how to calculate it. I'd chosen the 220uF simply because I have a few of these lying around and was using them across the opamp power and across my voltage divider. Its only job is to block DC, not give me a specific 3dB point. If, as you say, I can get much higher on the R values in the feedback network, I'll happily swap it for a 2uF poly cap. I'll play about with this. Thanks again. – Oli Comber Nov 27 '15 at 15:51
• Just to confirm - the TL064 is the one I'm actually using. I don't know much (yet) but from my little bit of reading it seemed to be a reasonable quality, low power, single supply, low cost part. More than happy to change it if anyone has any good suggestions. The image I pasted was from a new partsim project I started just to make that mock, and I didn't import the TL064 spice model. – Oli Comber Nov 27 '15 at 15:57
• See my extra thought on the TL064 – Andy aka Nov 27 '15 at 16:08
• Yep that makes sense - cheers. I think I'm ok here because I'm designing around amping a line level ~1v to 1.7v, so I've set the overall gain to about 1.5 here (accounting for the attenuation of the input). This should give me the flexibility to use many sources and have the waveform fill the whole available space - max volume! :-) The TDA7492 chip/board can only take 3.6v p-p (1.8v peak). So I'm fine with this limitation. When this runs outside, it will be on 24v batteries, and indoors on the 12v supply I only use it quietly - probably only a few millivolts. – Oli Comber Nov 27 '15 at 16:28
• Although before I can really decide on the gain to use, I need to test the line level and/or headphone voltage of a few devices - tablets, phones, my Pi and various cheap USB audio dongles all seem to vary wildly. – Oli Comber Nov 27 '15 at 16:30

While C1 is not in the straight line from source to output, it is obviously in the signal path. Consider, if you could vary the value of C1 at an audio frequency (constant voltage, not constant charge pedants, please, this is a non-physical variation), it would modulate the output signal by changing the frequency at which the feedback path rolled off, just as much as a coupling cap in the signal path would by exactly the same mechanism. In contrast, a power supply decoupling capacitor would not modulate the output signal if its value was varied.

The 6v battery you have drawn for bias may short circuit your simulation input, in case you run it and find you have no gain!

Don't take the golden-ears brigade too seriously about needing gold plated capacitors hewn from solid unaffordium, electrolytics are fine in the signal path for 99% of the population, 100% if it's a blind listening test! Beware the value of electrolytics though if you want an accurate filter corner frequency, they tend to have wide tolerances. In the position you have shown, a frequency down inth eHz makes that issue moot.

• +1, thanks v much! Don't worry too much about the very simple circuit I put in to the question - this was by request. But you'll see in Partsim (in link) that I don't have this problem, and it works fine. I've built a similar circuit to this before using electrolytic Tantalums for coupling and it sounded lousy. I've since found out about ESR, inductance and leakage. I like to learn this stuff properly, which is why I'm here and not some "golden ears audiophile board" :-) I'll take your advice and give it a go and see how it sounds. Thanks again :-) – Oli Comber Nov 27 '15 at 14:14

I think it is in signal path. Output directly depends of the value of the capacitor. Use ceramic capacitor.

Maybe try to put C1 directly connected to non-inverting input and R7 in parallel. Possibly you need to change the values for feedback.

• Interesting idea - I'll look at ceramics again. As has been pointed out, the lousy tolerance of ceramics wouldn't be a problem for this. Putting C1 and R2 in parallel would give me zero AC gain and much DC gain - the opposite of what I need. Thanks though, I'll try the ceramic tip :-) – Oli Comber Nov 27 '15 at 14:21