"
Mr Barnard: Shut your festering gob you tit! Your type makes me puke! You vacuous toffee-nosed malodorous pervert!!
Man Look! I came in here for an argument.
Mr Barnard (calmly) Oh! I'm sorry, this is abuse.
Man Oh I see, that explains it.
Mr Barnard No, you want room 12A next door.
"
From Monty Python's Flying Circus, the argument clinic sketch
Compared with school it's nothing . I should point out I worked on a farm and was very tough , it helped . One or two contributors are too serious and don't take criticism well . I say some daft things and get shot down in flames when I do . Sometimes that's when I learn most . I often know what I am saying is provocative , it's just good to go through the arguments one more time . For example and not me who said it . Voyager may not be responding to time in a way that we understand as it leaves the Solar System . I 99.9% don't think it worth repeating , 0.1% was enough that I wrote it . Do I beleive it ? 99.9 % no .
This is something from work . LM317 . Zero interest in noise when doing the test and no requirement for it to be low . Just being certain no oscillation . The references voltages are about safety and not typical use ( we call 230 V 300 V when we can ) . The 317 will see no more than 32 V and 25 typcial . A double 317 was used as it is cheaper than a heat sink . 2 x 1R used in outputs and them connected as if a single regulator . No extra 10 uF to the adjust pin . Usual 0.1 + 1 uF as per application notes ( one set for two ) . Output voltage 10.2 V .
What I think we see here is statistical noise cancellation ? Interestingly not only lower but nicer . I have used multiple parallel op amps for noise reduction in the past . Perhaps this is why I liked it ? Anyway , thought it worth a share . I think I was told 66.5 mV ( divide by root 2 ) would be the least possible using this technique and would be 3 devices . Doubtless another reason might be the real one ? The current was 1 x 110 mA or 2 x 55 mA .
To be expected from paralleling, to be sure. In principle no limit to that, and with statistical independence of the internal references, will give a reduction factor of the square root of N (the number of devices). The effect on line rejection is another matter. There may be other interactive effects given the low resistances used for the combination, and one could be a little concerned about differences in the voltages of the internal references if the aim was an assured noise reduction. But as you point out this was not the aim of the paralleling, merely a reduction in device dissipation.
I reran the tests with a DC source . Not much to tell , a small reduction in noise . I should have said 0.665 mV as suggested limit . As said previously it is the shape of the noise that seemed to have changed ? I thought about output impedance . With 2 x 1 R sharing resistors the change should be for the worse . I thought about the lower current , if I get a chance I will rerun that . I might try a quad op amp , see where that goes . I especailly like them as the pin out is symmetrical , less chance for errors . LM 324 is nice and noisy , it might give more to look at . TLO74 and MC 33079 also . For once I see no reason to extend it to high cost op amps . The trends should be universal . 4 x MC 33079 should give less than 2 nV / root Hz in paralell . Using one at a gain of 62 at 500 uV ( max ) seemed to sound excellent . From what I understand the electrons would be scarce at that level . It beggars belief sharing the electrons out between 3 op amps in my case can work ! It sounded better especially in terms of warmth . Seeing these graphs I feel the warmth is explained . I have said that transistor sound is VAS + crossover distortion , perhaps add noise shape ? Tim de Paravacini was saying this years ago , I get the feeling the journalists somehow mildly lost the point . It is not the lower noise , it is the noise shape perhaps ? The way it was said is if we lower the noise as far as we can everything gets better ( especailly 1/f ) . I really doubt that , in fact I can see how noise could be useful in small amounts ( white ) . Shape of noise must matter . With a noisy source perhaps it will be clearer . One test will be using gain to equalize the noise levels between paralell devices and one device . I suspect LM 324 the best choice for easy to read graphs .
One thing I did find is LM317 delivers on it's promises and costs peanuts . Mine is an industrial application , I almost feel ashamed to ask it to do this mundane work .
Naim Audio used 5 transistor in paralell for MC pick ups , this is when I first saw it used . I suggested recently someone use a 6 transistor long tail pair for a conventional power amplifier . Doubly so now that I have seen the nature of the noise .
One thing I did find is LM317 delivers on it's promises and costs peanuts . Mine is an industrial application , I almost feel ashamed to ask it to do this mundane work .
Naim Audio used 5 transistor in paralell for MC pick ups , this is when I first saw it used . I suggested recently someone use a 6 transistor long tail pair for a conventional power amplifier . Doubly so now that I have seen the nature of the noise .
I have a question for everybody, the more, the merrier.
How much and in what way does a CCS for the input stage(s) influence the sound?
From the resistor-zener-resistor basic to highly evolved two or three transistor CCSs. Since that subciruit is outside the NFB loop, a mistake with it is there to stay, as there is nothing to correct it. Theoretically, some rather evolved (complex) subciruits are capable of improving the performance, but I am not sure how, to what etxent (and what should be the acceptable minimum CMRR) and in what way is this audible.
Any views and thoughts are welcome.
How much and in what way does a CCS for the input stage(s) influence the sound?
From the resistor-zener-resistor basic to highly evolved two or three transistor CCSs. Since that subciruit is outside the NFB loop, a mistake with it is there to stay, as there is nothing to correct it. Theoretically, some rather evolved (complex) subciruits are capable of improving the performance, but I am not sure how, to what etxent (and what should be the acceptable minimum CMRR) and in what way is this audible.
Any views and thoughts are welcome.
Nige,
Some Japanese manufacturers have used as many as 6 parallelled pairs of FETs, followed by a bipolar Darlington stage, for their MC pre-preamps in a fully complementary mode.
You'll find many similar circuits to the one you mention from Naim in older Elektor issues. If memory serves, the BF series was widely used in Europe.
Some Japanese manufacturers have used as many as 6 parallelled pairs of FETs, followed by a bipolar Darlington stage, for their MC pre-preamps in a fully complementary mode.
You'll find many similar circuits to the one you mention from Naim in older Elektor issues. If memory serves, the BF series was widely used in Europe.
Some say use a clean PSU and resistor for amp input stages . Seeing as the input stage needs very little voltage swing there is scope for noise reduction in both hiss and hum .
I think Naim got the phono amp circuit from Wireless World . Bob Stuart told me a very interested mechanical engineer came to him after the lecture about distortion due to loss of information . Bob said he told him build a Sinclair amp , give it the largest power supply he could , get it reliable . Bob's final words to me was " I didn't expect him to put it into production " . The big deal is Bob never really came to market with what he suggested . Also he came late . As he said he was tied up with other things . It thus looked that Meridian copied Naim .
I think Naim got the phono amp circuit from Wireless World . Bob Stuart told me a very interested mechanical engineer came to him after the lecture about distortion due to loss of information . Bob said he told him build a Sinclair amp , give it the largest power supply he could , get it reliable . Bob's final words to me was " I didn't expect him to put it into production " . The big deal is Bob never really came to market with what he suggested . Also he came late . As he said he was tied up with other things . It thus looked that Meridian copied Naim .
I believe single ended input stages like the Naim Audio MC head amp introduce less noise than op.amp types with differential inputs.
With 5 parallel input transistors the resduction would be to a level a little less than half than the result of a single input transistor.
DVV has drawn a comparison of Naim's circuit with projects published in Elektor a long time ago. I remember one that used BFW16A in the place of the 5 *BC384 types in the Naim. The BFW16A is not an audio transistor but it has multiple emitters and the base spreading resistance is low.
Another approach used by Quad in the 34 preamp input is to use a singular NPN/PNP complement together as a single stage in push-pull to halve the noise.
I note that prime numbers feature in the foregoing selections.
I have tried stacking IC's op.amps - That did not work for me.
Michael J
Bingo! Michael, that's EXACTLY the one I was thinking of! Well done!
Yet another approach was used by Marantz in their 1978-1980 series of products. They used just one solitary FET as the only active component.
I'll dig in my log, I should have its schemati somewhere.
One story I heard was Philips demanded a chassis number for a TV if selling the transistors . It seems they were making ones as service repair items and loosing money doing it , these were the highly recommended ones . The Wireless World article championed certain power transistors as having lowest noise . We called it Rbb dash ( base spreading resistance I think ? ) .The Lentek pre pre amp used TO126 devices with the numbers shaved off . I suspect BF469/470 ? I loved the Lentek , never took to Audiolab their later offerings .
I was talking to my dad about this last night . He taught me electronics 52 years ago . I said I think the point people have missed is the cancellation is also a smoothing process . Thus the Himalayas become a more gentle if still high altitude range of hills . This gave us one hour of Quantum mechanics and why I shouldn't beleive it ( bless , he is 82 ! I told him Feynman proposed if Q M is correct a molecule is simultaneously anywhere and everywhere in the universe at any one time , my dad said he might accept that , I am sure that is urban myth and R F didn't say exactly that ? ) . Anyway , my ears say it sounds very different and lower noise is not a good enough explanation . This is the problem with electronics , the mechanisms are not understood as well as people think . Numbers or ears , soldering iron and eyes also .
I was talking to my dad about this last night . He taught me electronics 52 years ago . I said I think the point people have missed is the cancellation is also a smoothing process . Thus the Himalayas become a more gentle if still high altitude range of hills . This gave us one hour of Quantum mechanics and why I shouldn't beleive it ( bless , he is 82 ! I told him Feynman proposed if Q M is correct a molecule is simultaneously anywhere and everywhere in the universe at any one time , my dad said he might accept that , I am sure that is urban myth and R F didn't say exactly that ? ) . Anyway , my ears say it sounds very different and lower noise is not a good enough explanation . This is the problem with electronics , the mechanisms are not understood as well as people think . Numbers or ears , soldering iron and eyes also .
Nige, regarding smoothing, don't you do a rather good job of it if you stick in a say 10 Ohm resistor in series with the regulator output, with a say 2,200 uF capacitor to the ground after it?
It's big enough to act both as a filter at some 7 Hz AND a viable energy storage battery (assuming it supplies power to a low power consumption circuit)?
It's big enough to act both as a filter at some 7 Hz AND a viable energy storage battery (assuming it supplies power to a low power consumption circuit)?
That's very good point . The 10 uF I refer to is the one recommended on the application notes for reference to ground . It has a very big effect . The 1 uF they recommend I always took to be for stability . Reading more carefully it is to enhance the high frequency transient performance and doubtless does do a very good job of getting rid of the last bit of avoidable noise . The input 0.1 uF helps stability if not close to the reservoir cap ( if it is said not to be needed ) . I always fit both regardless . Douglas Self puts the output impedance of a regulator to be so low as to require Farads to match it .
I feel the 10 R and 2 200 uF would go nicely before the regulator with an additional 10 R . This would I feel be it's best position , it would prevent input short circuit . The point being that the LM317 is especially happy if doing less ripple reduction . 2 x 10 R shouldn't add to hiss too much . I would now without any hesitation use double regulators with 1 R output sharing resistors . I do suspect much of the regulator sound people claim to hear if true at all is simple noise shape ? Apart from that at audio the 1 uF is doing all the work . Please don't use tantalum , they always die in time .
I feel the 10 R and 2 200 uF would go nicely before the regulator with an additional 10 R . This would I feel be it's best position , it would prevent input short circuit . The point being that the LM317 is especially happy if doing less ripple reduction . 2 x 10 R shouldn't add to hiss too much . I would now without any hesitation use double regulators with 1 R output sharing resistors . I do suspect much of the regulator sound people claim to hear if true at all is simple noise shape ? Apart from that at audio the 1 uF is doing all the work . Please don't use tantalum , they always die in time .
- Status
- Not open for further replies.