LM3886 variation - filtered without el caps

[random007]

A variation that provides filtering for DC & ultrasonic/RF frequencies but doesn't use any electrolytic capacitors in the signal path. Should be a good compromise between signal quality and industrial robustness (though the gain is quite low - inevitable given sane component value restraints).

2nd chip amp project. Haven't built it yet.

Any thoughts welcome.

[Jay]

Originally Posted by random007

but doesn't use any electrolytic capacitors in the signal path.

What do you mean? C1 (obviously), C2, C3 even C8 (zobel) are in signal path...? EDIT: Oh, I see. Electrolytics. So 33uF MKP for feedback.

Originally Posted by random007

Should be a good compromise between signal quality and industrial robustness (though the gain is quite low - inevitable given sane component value restraints).

There was similar thread few days ago asking to compare Apex 3886 and the datasheet 3886. Why not choose any of them?

Originally Posted by random007

though the gain is quite low

The gain is approximately 82K/4k1=20x (not talking about the top freq), similar with the datasheet (20K/1K=20x if I'm not mistaken). The minimum is around 12x (so 12K/1K is possible with extra care with layout etc).

Tweaking with basic topology like this, there are points that I've learnt:

1) With equivalent capacitor compensation, the lower the gain the better the sound. Thus you have to make sure that everything (layout mainly) is good so that lower gain doesn't create instability (oscillation). You have to have tools to monitor this instability so you can use as low gain as possible.

2) In any amplifier circuit, the series resistor at input will affect THD. Because we cannot trust the accuracy of simulation, then tweaking by ears is still possible. In many opamps or even gainclone, I choose smaller resistance, not 1K. Type of this resistor is also audible. My taste is with carbon (Allen Bradley is cheap but never use this in feedback)

3) I have an hypothesis about phase issue and feedback. That is after building and listening (tweaking precisely) that lead me to CFB preference. I prefer (observation by tweaking by ears) smaller feedback resistor. The datasheet uses 20K. I can predict that 82K will sound worse (you need this big value so you can use smaller 33uF so you can use MKP). If you can go with 20K why 82K? 33uF MKP is expensive.

[AndrewT]

Then you go and ruin it with C4.

Nothing special, it just appears you have specified film caps for most of the capacitors in the build. = big cost and big space.

[macboy]

Originally Posted by AndrewT

Then you go and ruin it with C4.

Ahem. Really? An electrolytic on a logical input is going to "ruin it"?

[AndrewT]

the title pronounces

LM3886 variation - filtered without el caps

[random007]

In the title I said "filtered without electrolytic capacitors" : it would be in error to construe I meant the whole circuit was without them. I hope that clarifies the semantic aspect of this thread.

Not a very interesting thread, I know. There are a few designs that do DC without electrolytics but none I can see that filter both DC and RFI. Besides, I'm open to people more experienced than I (most people on this forum) making suggestions and/or correcting my calculations before I start soldering. As noted, it's a little more expensive to make with film capacitors, especially using reputable brands (not about to wade into the "capacitor brand = placebo" debate).

The two most common set-ups for the 3886 (etc) chip amps either omit bandpass filtering (the classic gainclone) or have a bipolar electro on the feedback because, at reference gain, the RC filter there needs to have a very high value cap. Dropping feedback drops the required anti-DC RC feedback capacitance to a level suitable for film capacitors.

The main reason for this attempt, trivial as it may be, is to remove the aforementioned bipolar electro from the circuit. Best practice in any circuit is to minimise the number of electrolytics as much as possible for reason of long term reliability. I'm sure I read an article ages ago by Rod Elliot (ESP) claiming bipolar electros significantly change capacitance over time. Of course, the bare minimum gainclone design requires almost no capacitors at all, but it is vulnerable to DC offset and RFI. Swings and roundabouts, as the saying goes.

[random007]

Originally Posted by Jay

<snip>

3) I have an hypothesis about phase issue and feedback. That is after building and listening (tweaking precisely) that lead me to CFB preference. I prefer (observation by tweaking by ears) smaller feedback resistor. The datasheet uses 20K. I can predict that 82K will sound worse (you need this big value so you can use smaller 33uF so you can use MKP). If you can go with 20K why 82K? 33uF MKP is expensive.

I'm going to have to recheck all my calculations. I chose 82K because a high pass filter on the feedback (matching impedance with the input?) requires either a large resistor or a large capacitor and I've already got some 33uF 100V film caps in my parts collection.

Thanks for the useful feedback (no pun intended).

[AndrewT]

I always suggest that the NFB roll off must not set the LF bandwidth of the amplifier.

Use the input filters to define the passband.

C1*R1 = 82ms.
C2*R5 = 140ms.
OK.

[DF96]

Originally Posted by Jay

In any amplifier circuit, the series resistor at input will affect THD.

That means that the amplifier circuit has non-linear input impedance - which can happen. However, if the input series resistor has to be small to avoid distortion then this means that the source impedance from which the signal comes must be even lower. If the chip is this sensitive to source impedance it really ought to have a buffer in front of it.

Originally Posted by Jay

I prefer (observation by tweaking by ears) smaller feedback resistor.

Smaller feedback resistors (other things being equal) will mean less HF boost due to stray capacitance at the feedback pin, but can also cause greater distortion due to thermal effects if the feedback resistor has highish tempco or too small power rating. A balance needs to be struck. This sort of thing can't be done by ear, as ears are very easily fooled. Much better to be done by datasheet, calculator and measurements; unless you are building an amp with built-in fixed tone controls and FX.

Finally, note that physically large caps can inject more hum and RF into sensitive circuit nodes so better screening may be needed than is necessary with the more usual electrolytics.

[Redshift187]

If you want to do away with electrolytics in the signal path, why not just use a servo?