Circuit topology vs parts selection vs sound quality

All the parts have to be put together at some point. Have fun!
Post Reply
gurevise
Posts: 9
Joined: Fri Feb 25, 2011 1:13 am

Circuit topology vs parts selection vs sound quality

Post by gurevise »

Hi guys,
I would like to discuss how to relate circuit design and components selection to the phono amplifier sound quality.

Lets start with this:
Would you agree that a high quiescent current topology sounds slower than a low current similar topology? I do not have a lot of experience with this but I read about it a lot. I always assumed that this is because of high current designs have low ratio of energy stored in bypass power capacitors to the energy needed to run active device (transistor or tube). Low current stages would run much longer on the same power supply energy and amplified signal is less affected by the power supply being modulated by gain stage energy demands.

Any comments will be greatly appreciated. If you disagree, I would appreciate if you explain your reasons why.

Thanks
Sergey
Dayton OH
USA
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

you said:
"Would you agree that a high quiescent current topology sounds slower than a low current similar topology?"

No. but think about your use of the word "slow". not a very descriptive adjective for what goes on in a generic voltage amplifier, with little or much current going through it. the word "sounds" is also truly problematic... what is it that you are actually wondering about?

and:
"I always assumed that this is because of high current designs have low ratio of energy stored in bypass power capacitors to the energy needed to run active device (transistor or tube)"

Why would that be? if it is a voltage amp and biased in class A, where the average current over a duty cycle is the same always... what would it matter if it was small constant current or high constant current? of course, if your generic voltage amp is sharing power with several other stages and has poor decoupling, problems can come up. but not from the causes you imagine.

you say you have read this somewhere... this wasn't romy the cat, was it? i'm suspicious about your use of the word "slow" and the implied opposite "fast". well, everything in a phono preamp affects performance and "sound". the simpler it is, the more the parts usually affect things.

lets get one thing out of the way. audio bandwidth is wide (ten octaves) but slow (20KHz max), in terms of electrical phenomenon. high frequency work generally requires higher currents than are "typical" for audio work because of the difficulties of parasitic shunt capacitance: at really high frequencies, wire is not required for the movement of currents because the energy of the signal can be capacitively coupled to anything nearby. video, radar, and computer circuits are typically loaded by 90, 75 or 50 ohms so that the C doesn't screw things up too much. the layout is critical to reduce/work with the surrounding environment. do you know how much C you need for a dead short at 10K ohms and 900 MHZ? it ain't much. but you try swinging 5 volts of linear ac at 900 MHz over 50 ohms... you will find you cannot easily do it with a class B voltage feedback op amp. not without lots of tricks and quite a bit more than the expected amount of current.

in a generic voltage amp, more likely is that all distortion is reduced and the bandwidth (speed?) increases as the quiescent current increases. quite often, the headroom is reduced, although in a preamp, this isn't usually the main issue. the "faster" your circuit needs to move, the higher the current pouring through the circuit. i'm not sure this is what you mean though? the "secret" to the crazy low distortion solid state stuff, like halcro or doug self's stuff is a HIGH current in the voltage amplifier stage. not only the insane amounts of negative feedback used. i suppose you could argue that these amps and preamps "sound" "slow". i wouldn't agree with you there. not my cup of tea, but certainly not slow.

another issue is noise. particularly for phono. there is an inverse relationship between amplifier generated noise and transconductance. all amplifier types rely on this to function. there are many issues involved but, basically, the lowest noise necessarily comes from the highest transconductance amplifier. this means that the quietest devices are transistors, fet and bipolar. that is because tubes cannot compete there. for tubes, the transconductance is more closely related to the operating current because there is so much less of it. more is more. tubes are also HOT, which sets a noise floor all by itself. the theoretical limit to noise in an ideal amp is 0.2nV per root Hz. that can contain several stages at this point in time. it has to do with the temperature of the planet, the background noise of this universe, semiconductor characteristics and that electricity requires a conductor with particular qualities. i believe the quietest phono stages measured are all in the 0.3nV - 0.4nV per root Hz range... halcro, john curl's various bits, and FEW others. that's the noise of a 5 ohm resistor. tubes can't do this. (a particular tube with a step up tranny can get close... but still not as quiet)

and as for the power supply components, yes they impart a character. electrolytic caps are famously "slow". actually, some of the charged polymer stuff made today is lower esr and less lossy at small currents than the "best" polypropelene/foil film caps made... even though they leak like a sieve at small levels (cartridge sized levels), which poly doesn't. they are not "slow" by any definition, as they are for really high frequency switching applications. i'm getting distracted. still, what does "slow" mean for a bypassed voltage amp? it means slow to discharge and/or recharge? it means they have a variable and parasitic resistance built in? that they change in characteristics slowly under power... what effect would that have? would it mean that the B+ voltage may shift following a large draw (excursion)? what would cause that? a large pulse or clip, or any low frequency continuous signal (bass). okay, an intermittent loss of low freqs, or a blunting of high slope pulse information... hmmm. this is based on the idea that the power supply would have less voltage/current available at those conditions? sounds plausible! anything else? maybe you're onto something?

the construction of electrolytic caps involves a roll of foils and insulators and a chemical electrolyte that stores ions... well, chemical processes take eons more time than electrical processes. long lengths of conductors have inductance, hysteresis, resistance. they are made out of aluminum typically... what's the "sound of aluminum"? (stop me). they have "resistance" so they must also make noise! ah yes... anything else? that noise is voltage variable... and the C is temperature variable. and the ripple current heats the cap... and...

all of this comes into play whether the design is LOW or HIGH current. and if the circuit is high gain and high impedance (low current) the time it takes to recover and return to 0 is LONGER. The noise is HIGHER. small parasitic storage mechanisms are more easily loaded making things MORE variable.
man, there sure are a lot of moving targets involved! why do we use analog components at all? i'm switching to digital!

perhaps, there is just something inherently "bad" about electrolytic caps? they make high current voltage amplifiers "sound slow".

yes, it's a ridiculous thing to say. audiophiles say things like this without thinking all the time. they will continue to do it. it will mean equally little when they do. no matter how many times it's said, it still won't mean anything more. if you know what you are working with, you can choose to work with it in a sensitive way. that is the work of analog. that is what one does.

for your information:

film caps work their worst at very small signal levels. this is because the parasitic losses and failings of the materials have relatively fixed levels (small and inevitable). dry film caps are much much simpler in this regard than wet film or electrolytic caps. this doesn't mean any one is better or worse "sounding" necessarily. dry film is simpler. but it all depends on the application. for large swings, film caps outperform inductors in nearly all parameters. less distortion, less losses of all kinds. in atomic work, one sees quartz capacitors that can release 20,000 joules of high voltage in nanoseconds into a low impedance load. film caps do work. just not at their best with 100uV signals. (that's why we like transformers... even though they have problems too) the dielectric absorption of some insulators alone is significantly, although fractionally, comparable to the signal at that level.

electrolytic caps have problems at very small AND very large changes of state. this is because the materials have parasitic effects that are of the same magnitude as the signal or "fail" at these extremes. in the middle, there is a really useful window of application.

and finally, some of these effects sound GOOD. if you are making a guitar amp, you probably will want to try all sorts of things you would never try in a phono preamp.

i am not qualified to suggest how certain parts "sound" except i will say that carbon comp resistors are friggin noisy at small cartridge sized signal levels. that's why metal film and bulk foil and wire wound are "better" in the front of a phono voltage amp. i have tried to live with em but i really don't like noise. build something up and let us know how you think it sounds. lots of help here to do that...
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

you know, i have to apologize for my last post. i was jet lagged, ornery, and i don't like audiophilia. weird that i am surrounded by it. anyway, i shouldn't have jumped on this the way i did. let me make it up to you?

i really have to watch myself when i get around the word "sound", in an audiophile context. it is a meaningless word. it has been since harry pearson actually... not his fault, but he had the hammer in his hand when it happened.

lets do an experiment, and thanks to the print screen button and LT spice, we have a convenient way to convey meaning to each other.


exhibit A: i have sketched up 3 slightly different arrangements of a grounded cathode 12AX7 gain stage. all have the same bias (-1.4 volts), the same peak input signal (1 volt AC), the same plate voltage, and necessarily, roughly the same plate current. each stage is loaded differently. one, is loaded with a resistor with a value about equal to the plate resistance, 1:1 loading. two, is 2Rp. and 3 is constant current. i have arranged it fixed bias, just to keep things simple.

LT spice generates very believable output waveforms (from the plate) which i have superimposed on top of one another... the smallest amplitude sine wave, the one with the most distortion, almost all of which is on one side of the plate output, is the 56k/1:1 loading. the middle sized sinewave, that looks slightly pinheaded but much better than the 1:1, is the 2:1. 2Rp. and third, the waveform with the biggest amplitude, is also the one with the least distortion... that's the constant current load.

let's have a look.
Attachments
waveforms.gif
waveforms.gif (35.76 KiB) Viewed 6080 times
gain stages.gif
gain stages.gif (223.11 KiB) Viewed 6080 times
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

okay, now lets look at the fft plots of the various "topologies" (they are all the same except for load).

first is the 1:1. 2nd, the 2:1 and last is infinity:1
Attachments
CCS fft.gif
CCS fft.gif (34.8 KiB) Viewed 6079 times
120k fft.gif
120k fft.gif (38.08 KiB) Viewed 6079 times
56k fft.gif
56k fft.gif (37.18 KiB) Viewed 6079 times
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

ahh, i've screwed up already... the middle one is 4Rp. that's ok. it's the trend that matters.

okay now comes the gain/bandwidth info... obviously this is into an infinite load (one stage and no resistor or tube following) so bandwidth is much better than what you would ever see in real life. but just keep your eye on the trend.

the constant current source loaded has the most gain: 39.4 dB. -3dB at 465KHz. the 4Rp stage is next with 36.8 dB, -3dB at 563KHz. the 1:1 loading is lower with 31.9 dB and -3dB at 1.1MHz.
Attachments
bandwidth info.gif
bandwidth info.gif (30.71 KiB) Viewed 6078 times
my skin is thin, my skull is thick.
dave slagle
Posts: 2085
Joined: Sat Jan 22, 2005 3:54 am
Location: NYC
Contact:

Post by dave slagle »

What I notice about the last picture is that the final 6dB slopes of them all overlap. Gotta get my head around what this means....

dave
Get Your Fix
www.hifiheroin.com
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

ok, a few comments. clearly, the 1:1 loading sucks for hifi when you are plying it with lots of signal (which is the case here, intentionally). the fft reveals something classic and expected. as the oncoming of clipping approaches, each harmonic rises in a characteristic common to most triodes in varying degrees. at regular octave steps, each successive harmonic component of the output is less than the one before it... down to a level where the higher order stuff blends into the noise floor. at this particular input level, this stage only makes it down to the 5th harmonic before bottoming out... but its a nice even fade. what does this "sound" like? 20% distortion... it "sounds" thick, with brighter overtones and somewhat crunchy. pretty much like john lennon singing though a distorted mic pre, which he nearly always did because he hated his voice clean. not good for hifi, but actually not bad for some other things... actually, because this arrangement has 2 times the bandwidth of the other 2, if you didn't need a big swing, and speed was your thing, this might still be useful for something.

the 4Rp loaded one also shows some classic stuff: the 2nd harmonic is -40dB down from the output. the 3rd harmonic is -70dB down. this is why the 12AX7 is the worlds most popular tube. note, however, that the higher order stuff is higher than the 3rd! hmmm... what's up with that? how does this "sound". slow? well, one thing is that its much much cleaner than the 1:1. and in fact, this operating point is probably the most widely listened to operating point of vintage hifi... start with heathkit and work your way through to marantz. 660uA and 150 volts is so many preamps and guitar amps i couldn't count. you have heard this sound before. with and without feedback error correction.

the last one is more modern. the CCS loaded stage has the "best" performance in terms of gain and distortion. it also has the poorest bandwidth... the "slowest". here, in these examples, the same current, voltage, bias and signal level have been maintained. only loading has been changed.

tomorrow, we can try to vary the current but maintain the same load... or maybe you can do it?
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

its the miller C
my skin is thin, my skull is thick.
dave slagle
Posts: 2085
Joined: Sat Jan 22, 2005 3:54 am
Location: NYC
Contact:

Post by dave slagle »

nanana wrote:its the miller C
I can see the plate to ground C causing the lowest output impedance source (1:1 load) to have the widest bandwidth but I haven't seen the final rolloff slopes overlap before like they do in your sim.

dave
Get Your Fix
www.hifiheroin.com
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

but, in my over simplified scheme, there is one fixed C value, no amplifier following (no C serial "multiplication"), the only thing changing is gain. so the miller C is the "dominant pole". you have stared at this before... think of open loop op amp charts.

think of feedback gain reduction...
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

the miller C is between the grid and plate...
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

so here is the next step, the 2nd step in 12... for those who came to believe a power greater than themselves (LT SPICE) could restore them to (audio) sanity.

here, the load is the same, the bias is shifted in order to draw more current through the tube. the signal has been reduced to 0.1 VAC peak so it doesn't clip (the overhead is reduced) the higher current arrangements. the right most circuit is the same 4Rp 660uA, 150 Ep arrangement from the first step.

the first step was: admitted we were powerless (over audiophilia) and asked for help...

okay... now, everything has a similar look and the differences are smaller. this makes sense as the load is the same, the B+ is the same, we are only changing the bias. the plate voltage varies the most. 75 VDC with the "high" current bias, 100 VDC with the middling one, and the already posted "low" current setting with 150 plate volts. the plate currents are: 1mA, 880uA, 660uA.

with 0.1 v of signal, the gains are shown in the gain/bandwidth plot. note that the highest current bias gives the most gain AND the most bandwidth, for the same load. its not a huge difference, but it is a shift. why? transconductance. the paltry 1600 uMHOs of Gm available in a 12AX7 maxes out at 3mA, and drops to nothing at 25uA. at 1 mA, you have quite a bit more Gm to work with than you do at 660uA, even with a 12AX7, at these operating points. now the next thing is too much posting so i'm going to summarize from the fft plots from this second step.

"high" current: 2nd harm. dist. -45dB 3rd harm. dist. -52 dB high order floor -37dB.
"middling" current: 2nd harm. dist. -53dB 3rd harm. dist. -53 dB high order floor -36dB.
"low" current: 2nd harm. dist. -47dB 3rd harm. dist. -52 dB high order floor -35dB.

you can see from this that as current increases, the high order dist and noise steadily decreases (more Gm) and the distortion spectrum slides around with small variations.
Attachments
hi current fft.gif
hi current fft.gif (33.27 KiB) Viewed 6046 times
waveforms 2.gif
waveforms 2.gif (30.16 KiB) Viewed 6046 times
bandwidth info2.gif
bandwidth info2.gif (31.83 KiB) Viewed 6046 times
gainstages 2.gif
gainstages 2.gif (78.16 KiB) Viewed 6046 times
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

ok, so this next test has the 12AX7 biased the same with 3 different loads. in this case we have an inductor, the 4Rp loaded stage and an 8Rp loaded stage. i have adjusted the B+ to allow the same current through each stage: 1mA. i have intentionally undersized the inductor (it would need more inductance to be "properly" loaded for a hifi app) so that you can see the negative issues of inductors more clearly, as well as the benefits. my purpose here is to show clearly that there is no free lunch. no "best". no. someone tells you that, they are delusional. lying to you, themselves or both. 144 perfect ways to do anything analog. anyway, the input level here is 0.1 volt peak (100mV). the operating point is maximum intermittent dissipation (2.5 watts); 250 plate volts and 1mA plate current.

what you see here is that the choke loaded stage has the same gain, practically, as the constant current source loading. however, several important differences are obvious: the loss of loading at low frequencies means the bass rolls off, and the choke loaded waveform is shifted IN TIME in comparison to the resistor loaded stages, and it requires the least in terms of B+ and yet has the biggest swing of all 3. how can it swing up to 261 volts when its B+ is only 252 volts? because the energy stored in the magnetic circuit is stored and released between B+ and the plate resistance of the 12AX7. the problem is that it takes time to charge and release this energy. so the choke load has many advantages. it reduces the B+ requirement, increases the gain, and allows swings above the B+ rail (a big potential advantage in terms of clipping performance...). but it does all this with a penalty... it adds time to the arrangement, beyond what would be expected from the signal source itself. there are more subtle and complicated issues connected to this: the time added is not completely linear or evenly delayed. there are parasitic storage and loss mechanisms in a choke which hang on longer or more completely at certain frequencies than others.

resistors are much simpler electrically, and this simplicity is a great advantage. a look at the waveforms show that even if the load is doubled, time remains in the same relationship it did with the signal source. a disadvantage becomes apparent also... in order to maintain the same plate voltage/current ratio while increasing the value of the load, the B+ has to be increased to an astounding level! the 470k plate load needs a 700+ volt supply to keep 250 at the plate at 1 mA. ahoooah! all of a sudden , the time penalty of the choke loaded stage seems ridiculous to making a stage that might kill you or your pets/loved ones in a short unfortunate moment. why would we need to have such an operating point? good question.
Attachments
gain bandwidth info 3.gif
gain bandwidth info 3.gif (31.22 KiB) Viewed 5989 times
waveform 3.gif
waveform 3.gif (37.36 KiB) Viewed 5989 times
gainstage 3.gif
gainstage 3.gif (202.46 KiB) Viewed 5989 times
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

here are the waveforms and fft plots from the 3 when i increase the input level to 1.5 volts peak. note the large beautiful swings, from a lowly 12AX7. this is why one might need such a large supply. a large swing. all power triodes need these levels of swing at their inputs. for 1.5 peak volts in we have a swing between 410 volts and 115 volts for the choke loaded stage. note the wave form is not perfectly symmetrical any more. distortion is happening somewhere... but the swing is HUGE. the resistively loaded stages are more symmetrical, but lower in amplitude. a look at the fft's says it all.

the choke loaded stage has the most gain, +40.5 dB, but the highest distortion... 2nd harmonic is +5 dB and 3rd is -6.7 dB. the 220k load gave a +38 dB with -4 dB and -18 dB respectively. and the 470k load gave +39 dB with -9 dB and -33 dB. clearly, the choke is beating its head against B+ rail and the varying Rp of the 12AX7.

no free lunch. so up until a point, the choke load really shined even with it's added time issues, but at some point, the asymmetry of the load points (B+ and 12AX7 Rp) causes non linearity. one is fixed low and the other is high impedance. the lowly resistor starts coming into its own again.

okay, this admittedly oversimplification was designed to show some overlapping and contradictory concerns. more coming.
Attachments
big swing fft3.gif
big swing fft3.gif (32.79 KiB) Viewed 5985 times
big swing fft2.gif
big swing fft2.gif (32.97 KiB) Viewed 5985 times
big swing fft1.gif
big swing fft1.gif (28.75 KiB) Viewed 5985 times
waveform 4.gif
waveform 4.gif (36.95 KiB) Viewed 5985 times
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

okay, in this next step, we get more sophisticated... still highly idealized but trend seeking even more so.

here, we have the ordinary 220k resistor pitted up against its evil nemesis: the srpp stage. also elbowing its rough mannered rudeness into the fray, the sallowed fetid mu follower. and not to be ignored, the ccs loaded stage keeps its warty foot in the door.

notes: there is VERY little difference in gain between the 200k resistor and the srpp stage... obviously the srpp is neither "constant current" or higher performing in terms of distortion... the fft is very close! hmmmm...

also, and maybe not so surprising, the mu follower's gain is almost identical to the ccs stage. the fft's are different, but the ccs stage is so ideal, and the mu follower is modeled somewhat less so (good). clearly the muF is nearly a ccs load for the lower half, unlike the srpp poser. however, the distortion is still higher than for an ideal simple ccs.

okay. gloves off. now, i will actually make these wimpy fakers do some work... and we'll see what they are actually capable of doing.
Attachments
waveforms gs4.gif
waveforms gs4.gif (37.53 KiB) Viewed 5962 times
my skin is thin, my skull is thick.
nanana
Posts: 143
Joined: Mon Jun 14, 2010 7:34 pm
Location: malmö sweden

Post by nanana »

ooops. the schemat didn't make it. the super crappy internet i have (sweden sucks for it) strikes again...
Attachments
gainstage 4.gif
gainstage 4.gif (245.14 KiB) Viewed 5961 times
my skin is thin, my skull is thick.
Post Reply