I folks,
It's my first post on this forum I'm very interested in amp design and I would like to submit this subject (kind of mystery) to you.
I recently discovered the concept of voltage coefficient in resistors: The resistor value decreases with voltage applied.
The change in resistor value is small: metal film according to datasheets can have values as small as 1ppm/V in 0.1% range but most metal film have voltage coef much above 5ppm/V.
Now imagine you have designed the perfect 100W/8Ohm audio amp with 0% THD in open loop (!!!).
You introduce your FB loop using your tremendous (for example RC55Y) part that has only 1ppm/V of VCR and then ...
Considering you get approx 40V peak voltage on the FB resistor : your FB resistor is affected of a 40.10-6 fluctuation factor. Which will give a closed loop DHT of around 0.004%.
OK It is small but it is much more than some figures I've often red.
Can some one help understand this strange fact ?
Regards
Stéphane
It's my first post on this forum I'm very interested in amp design and I would like to submit this subject (kind of mystery) to you.
I recently discovered the concept of voltage coefficient in resistors: The resistor value decreases with voltage applied.
The change in resistor value is small: metal film according to datasheets can have values as small as 1ppm/V in 0.1% range but most metal film have voltage coef much above 5ppm/V.
Now imagine you have designed the perfect 100W/8Ohm audio amp with 0% THD in open loop (!!!).
You introduce your FB loop using your tremendous (for example RC55Y) part that has only 1ppm/V of VCR and then ...
Considering you get approx 40V peak voltage on the FB resistor : your FB resistor is affected of a 40.10-6 fluctuation factor. Which will give a closed loop DHT of around 0.004%.
OK It is small but it is much more than some figures I've often red.
Can some one help understand this strange fact ?
Regards
Stéphane
The NFB resistor fluctuates (max) by 40e-6 ppm, which you forgot to divide by 40 to get the ratio, so it's (40e-6/40)*100 in %=0.0001%
Next, these 40e-6 ppm are the peak value when the sine reaches Vpeak; 99% of the time this value will be smaller or even zero. THD measurement averages these values and will give a further reduced reading.
Last, this number will be reduced by negative feedback just like any other error within the amp as well.
Your example sounds like a low NFB design (or big power amp), but these have still 20-30dB NFB which reduces this number by at least another factor 10.
Have fun, Hannes
Next, these 40e-6 ppm are the peak value when the sine reaches Vpeak; 99% of the time this value will be smaller or even zero. THD measurement averages these values and will give a further reduced reading.
Last, this number will be reduced by negative feedback just like any other error within the amp as well.
Your example sounds like a low NFB design (or big power amp), but these have still 20-30dB NFB which reduces this number by at least another factor 10.
Have fun, Hannes
Hi Hannes,
I not exactly agree with your analysis, I did not forgot to divide by 40 ...
40ppm = 0.004% is the actual fluctuation of NFB resistor (if we consider the voltage on negative input as small against output voltage).
It is not a voltage error against output voltage.
It is the modifcation of NFB resistor during a signal period.
If closed loop gain of the amplifier is G in a non inverting configuration. the simplified NFB loop is a voltage divider made of said NFB resistor plus an other R Resistor. The gain of the amplifier when open loop gain is big against G is:
G = 1 + Rnfb/R
So we clearly see that a 0.004 % modification in Rnfb is almost a 0.004% change in G which means a 0.004% DHT.
This cannot be cancelled by feed back while feed back is the core of the issue.
Regards,
Stéphane
I not exactly agree with your analysis, I did not forgot to divide by 40 ...
40ppm = 0.004% is the actual fluctuation of NFB resistor (if we consider the voltage on negative input as small against output voltage).
It is not a voltage error against output voltage.
It is the modifcation of NFB resistor during a signal period.
If closed loop gain of the amplifier is G in a non inverting configuration. the simplified NFB loop is a voltage divider made of said NFB resistor plus an other R Resistor. The gain of the amplifier when open loop gain is big against G is:
G = 1 + Rnfb/R
So we clearly see that a 0.004 % modification in Rnfb is almost a 0.004% change in G which means a 0.004% DHT.
This cannot be cancelled by feed back while feed back is the core of the issue.
Regards,
Stéphane
I don't understand why you want a factor of 40. I agree that if the amp has 6 dB noise gain (i.e. equal shunt and feedback resistor values) the voltage coefficient will cancel (given resistor value and voltage coefficient match). However for the usual noise gains > 20 dB there is almost no cancellation and voltage coefficient will proportionally affect linearity.
By definition imperfections of the feedback network will not be reduced by loop gain. If you put a diode in your feedback network the output will be distorted. Otherwise we couldn't make precision rectifiers and clamps using opamps...
Voltage coefficient is extremely difficult to measure down to 1 ppm/V. 1 ppm/V is likely to represent the measurement residual rather than resistor imperfection of decent metal film parts. I've investigated this myself using special distortion measurement techniques and I've not been able to find distortion above -150 dB at +20 dBu. Surprisingly the $20 Vishay bulk metal parts are actually worse with this respect than most standard metal film parts I've measured...
Of course voltage coefficient *is* a serious problem for carbon and thin/thick film resistors. Another issue (for any resistor except those with near-zero tempco) is distortion from thermal self-modulation at low frequencies. I'm doing some research here at the moment but no usuable results yet.
Samuel
thermal and voltage ratings
is this why some suggest that peak voltage use <10% of the resistor's power rating for the NFB loop?
is this why some suggest that peak voltage use <10% of the resistor's power rating for the NFB loop?
There's another discussion about distortion and excess noise in resistors here: http://www.diyaudio.com/forums/anal...curls-blowtorch-preamplifier-part-ii-393.html
There's some surprising results with distortion measurements there too.
I expect that's more to do with temperature coefficient - the increase in resistance with self-heating.
To reduce distortion, perhaps one could connect several resistors in series, so that each one has a smaller voltage across it?
(Not that I expect to design a circuit good enough to tell the difference anytime soon😛)
Last edited:
Can't finde the details how these measurements are done; anyone knows the page number..?
Well, that won't address any voltage coefficient issues (rather the opposite: most high-power parts actually have higher voltage coefficient) but at least the thermal self-modulation I mentioned because higher power parts typically have slower thermal time constants which moves the onset of this distortion mechanisms towards lower frequencies.
Sure. But as I said: It ain't a serious issue anyway.
Samuel
Some literature: http://www.barthelectronics.com/pdf...1 Voltage Coefficient Products_Pulse Page.pdf
Samuel
Samuel
an extreme solution is to use the same value resistor in your feedback network - need a gain of 20 - use 20 identical 1K resistors
then any matched common distortion mechanisms such as V or T coefficients do not effect the division ratio
then any matched common distortion mechanisms such as V or T coefficients do not effect the division ratio
JCX's solution is highly recommended in my view.
I've changed my feedback resistors yesterday.
Resistors was 1k and 2k, now I use 2k to (2k+2k) in my headphone amp. 2k was used, beacuse these was at home currently.
I know that the dissipation on each resistor has halfed (1k changed to 2k), but it's more important, that (as JCX sad) the voltage ratio is now constant at any voltage level ! No more distortion from different thermal characteristic and no more distortion from voltage coeffcients induced ratio change.
Listening impressions:
I hear a little more resolution, better transients. Better highs. 🙂
My own made headphone amplifier is a really good amp, and maybe you will not hear any change in your amp. Try it in a high gain amplifier, where the ratio is higher and the errors ( coming from dissipation and voltage coefficients ) are higher too.
It's worth it, and not costs more than 1 USD.
Many thanks JCX !
Greetings from Budapest - Hungary!
A good idea from jcx (if the issue is significant) but am I the only one who senses a new audio fashion coming on? Forget damping factor, DC-daylight bandwidth, and power station sized reservoir caps: the latest must-have thing will be identical resistor feedback. We need a snazzy name for it: how about 'replicated feedback string'? Everyone will 'hear' how much it improves things. It is the ideal audiophile upgrade: sound physics behind it, yet the improvement will be virtually unmeasurable in almost all amps. You can't argue with that!
Hi Stéphane !
Many Thanks for thread starting !
Your calculations are right. Try JCX's solution. It's working for me.
I feel, that the problem is with dissipated power, not with voltage coefficients.
The feedback ratio changes moment by moment because of different resistor surface temperature.
But if You make your feedback from the same resistors, the surface temperature (hence resistance values too) will change equally so the ratio will not change.
Best regards,
Krisztian Aklan
Ed Simon published some interesting tests on resistors and their contribution to THD in Linear Audio. I believe that is what was discussed in the Blowtorch thread...
_-_-bear
_-_-bear
I saw your webpage and it's well written. I hope You know that THD is irrelevant when speaking about transients (music). Even an IMD measurement is a static measurement. Could you show me a music, that has a static spectrum ? ( Of course you can't. ) Only measuring signals have static spectrum.
So I don't wan't to measure it. It's enough if can I hear it.
The change in sound is significant so grab your soldering iron and try it yourself ! 🙂
Regards,
Krisztian
Don't you want to measure it rather than speculate about it ?
Solder them and measure the resultant inductance. This will shortly end this conversation. 🙂
You don't know have resistors are made with spiraling ???
You forgot, that lower resistance resistors have lower inductance. So I bet on, that the inductance of 20 pieces 1k resistors will be on the range of one piece 1k resistor's inductance. Maybe it will 2 or 3 times more than 1k resistor's inductance, but not 20 times more as you think.
greetings,
Krisztián
Are you a follower of Douglas Delf ? What a shame.
So the equations on your webpage are so simple, that they're so far from real world. You may try to write some differential equations for momentary dissipation of transistors, try to include dielectric absorption of capacitors, and try to model resistors (voltage coefficient, tempco).
All of these effects will lead to complex differential equations.
Made some simple calculations in Excel with 15 volt applied to feedback resistors. The effect of voltage coefficient with ultra precision (and ultra pricey) Vishay Z201 resistor (one pcs 1k and one pcs 2k) will be the same as with 3 pcs 1k standard 1% resistors ...
The effect of voltage coefficient caused division ratio change is compression of music.
You can eliminate this compression if you build your feedback from equal resistances.
See the attached picture and don't waste your time with modeling. Listen some music !
Attachments
- Status
- Not open for further replies.