Feedback Resistor Values

[Puffin]

I have built several Gainclones. The most recent being an LM3875 (build details on Decibel Dungeon) as I wanted to get as close to the original Gaincard as possible)

DecDun uses 2 423K resistors in Parallel to get about 211k (used in parallel to increase the power rating). I in advertantly wired them in series and got massive gain, but something I found very likeable. I corrected my mistake and the amp sounded great, but a little poilte in comparison. I have put some connectors in so that I can swap the resistors in and out easliy.

The DecDun LM3886 uses 270k

At the moment I am using 390k whcih I think sounds great.

There is so much variety of values I have seen on different layouts for the both the LM3886 and LM3875, 22k, 100k, 47k etc

I am learning quickly, but wondered if anyone could tell me why some specify low values, and some high ?

Is there an optimum, or is it personal preference.

[gootee]

I didn't go and try to find the circuit you actually used, since you didn't provide a link. (Sorry. Too lazy right now, I guess.)

It depends a lot on whether the chipamp is being used in non-inverting mode, or not. Non-inverting mode usually has an automatically-high input impedance, regardless of the feedback R values, since the feedback doesn't connect to the input in that case. If that were not a consideration, then lower-valued resistors (within reason) might always be preferred, since they generate less noise of their own, and don't interact with stray capacitances and inductances as much.

There is, no doubt, an optimum for each topology and implementation. You could probably start by trying to determine the optimum gain for your setup. Look at the graphs in the chipamp's datasheet, at national.com, for example.

By the way: Why try to get as close to the original Gaincard as possible? It was not a wonderful design.

[Nuuk]

As Tom said, first consideration will be is it an inverted or non-inverted design.

Next, you need to pick the feedback and input resistor combination to keep the gain 'acceptable', ie in the SOA as specified on the datasheet for the particular chip.

Then, keeping the above in mind, you may wish to optimize the gain for your particular hi-fi system.

But it is interesting how changing just the feedback resistor can alter the sound. Is there an optimum value? Probably but it may be different for each builder and system!

[peranders]

Originally posted by gootee
If that were not a consideration, then lower-valued resistors (within reason) might always be preferred, since they generate less noise of their own, and don't interact with stray capacitances and inductances as much.

Indeed! I want a "pure" voltage divider in the feedback as possible. My design goal is to not let the feedback resistors not get too hot.

[Nuuk]

My design goal is to not let the feedback resistors not get too hot.

Isn't that the reason for using two in parallel?

[ttan98]

Originally posted by Puffin
I have built several Gainclones. The most recent being an LM3875 (build details on Decibel Dungeon) as I wanted to get as close to the original Gaincard as possible)

DecDun uses 2 423K resistors in Parallel to get about 211k (used in parallel to increase the power rating). I in advertantly wired them in series and got massive gain, but something I found very likeable. I corrected my mistake and the amp sounded great, but a little poilte in comparison. I have put some connectors in so that I can swap the resistors in and out easliy.

The DecDun LM3886 uses 270k

At the moment I am using 390k whcih I think sounds great.

There is so much variety of values I have seen on different layouts for the both the LM3886 and LM3875, 22k, 100k, 47k etc

I am learning quickly, but wondered if anyone could tell me why some specify low values, and some high ?

Is there an optimum, or is it personal preference.

The reason for using 680ohm and 22k combo is the 680 ohm determines the noise level of the Lm3875, the lower the value the lower the noise floor. You cannot pick too low, obviously there is a compromise, 680ohm is chosen. you easily use 560ohm or even 820ohm. After you pick the 680ohm you choose the gain you need, hence 22K is calculated. you could use 6.8K and 220K combo but the noise figure from 6.8k is too high, this combo is appropriately used as a power amplifier not as an integrated. I personnally would not pick 6.8K in either case. I use the 680ohm and 22k combo.

Sometimes when you listen to a high gain amplifier it seemed/perceived to sound better, infact you are hearing is that you are hearing the high gain amplifier set at higher gain compared the lower gain amplifier. You need to set both of them at similar gain to do comparison. I strongly suspect both will sound the same.

I hope I explain this sufficiently clear enough.

[Puffin]

Thanks for replies. I attach a copy of the circuit i used. ttan98, I am not sure where the 680R resistor fits in relation to this circuit, but I understand what you are saying.

http://myweb.tiscali.co.uk/nuukspot/...ecircuits.html

[Nuuk]

The 680R would be used in a non-inverted design Puffin. The one you have linked to is inverted. Inverted needs a much higher value input resistor as that plays a part in keeping the input impedance sufficiently high enough!

[Puffin]

Nuuk. Yup, that it explains it. Can I also ask. You do not show a Zobel network in your schematic. Is it not neccessary in this application?

Is it also worth experimenting with a Thiele network ?

[Nuuk]

That schematic was the first circuit that I tried and based pretty much on the minimum components that you can get away with. It worked well with my efficient speakers but I now include a Zobel on my current GC's 'just in case'.

I did try a Thiele network in my early experiments but convinced myself that it didn't quite sound as good with, as without, so I no longer use one.