I believe we can easily be misled into forming premature conclusions, the question is how many others do?
So lets say a speaker produces 5% distortion whereas an amplifier produces 0.01% (which I believe are typical figures). Why do people worry about amplifier distortion when half the time their speakers and room are producing hundreds of times the level of distortion of the amp?
Yet do you still believe that your conclusions are valid? It seems you do and I cant understand why.
Bala,
A bit of thinking out loud: the pivotal consideration with an audio amp is fixing the gain at less than 0dB at the high frequency pole. This requires a Cdom across the input and output of the VAS device. FACT: a typical HF pole for an audio amp is at 750KHz, orders of magnitude higher than the highest audible frequency. I believe you allude to this of course.
However, another FACT: the subjective listening experience is profoundly affected by the value of Cdom. Too low, and the amp suffers transient instability, lapsing briefly into small oscillations under audio provocation. Too high, and the gain at the pole is way below unity, and the amp is leaden, and slow, because the charge/discharge cycle of the Cdom starts to limit due to current starvation on the base side of the equation.
Like you, I've always been surprised that a few pF either way should have such a profound effect on sonics, but it does. Let us postulate a little.
The depletion layer in a working bipolar transistor is the analog of dielectric in a cap. The reverse biased pn junction of the base collector in a common emitter amp forms this depletion layer, which moves around due to charge migration, Vce, Vbe, etc. I have found repeatedly that faster transistors in this role make a difference; top end has more clarity. This defies the physics, at least to me. A 100MHz device sounds heaps better than a 10Mhz device, but moving beyond about 150MHz seems to bring little improvement. Little of this is visible on a CRO, but to me at least, it's highly audible.
A small extrapolation would indicate that a VAS with low Cob would then permit use of a slightly larger, and more linear, Cdom. The depletion capacitance is highly non-linear, and incredibly leaky besides since all the charge migration due to collector current flows right through it. Would this be a logical conclusion?
If not, do please let me know your thoughts, either here or privately through the website!
Hugh
A bit of thinking out loud: the pivotal consideration with an audio amp is fixing the gain at less than 0dB at the high frequency pole. This requires a Cdom across the input and output of the VAS device. FACT: a typical HF pole for an audio amp is at 750KHz, orders of magnitude higher than the highest audible frequency. I believe you allude to this of course.
However, another FACT: the subjective listening experience is profoundly affected by the value of Cdom. Too low, and the amp suffers transient instability, lapsing briefly into small oscillations under audio provocation. Too high, and the gain at the pole is way below unity, and the amp is leaden, and slow, because the charge/discharge cycle of the Cdom starts to limit due to current starvation on the base side of the equation.
Like you, I've always been surprised that a few pF either way should have such a profound effect on sonics, but it does. Let us postulate a little.
The depletion layer in a working bipolar transistor is the analog of dielectric in a cap. The reverse biased pn junction of the base collector in a common emitter amp forms this depletion layer, which moves around due to charge migration, Vce, Vbe, etc. I have found repeatedly that faster transistors in this role make a difference; top end has more clarity. This defies the physics, at least to me. A 100MHz device sounds heaps better than a 10Mhz device, but moving beyond about 150MHz seems to bring little improvement. Little of this is visible on a CRO, but to me at least, it's highly audible.
A small extrapolation would indicate that a VAS with low Cob would then permit use of a slightly larger, and more linear, Cdom. The depletion capacitance is highly non-linear, and incredibly leaky besides since all the charge migration due to collector current flows right through it. Would this be a logical conclusion?
If not, do please let me know your thoughts, either here or privately through the website!
Hugh
Last edited:
Yes it is quite possible to form premature conclusions, and there will always be those that say if you can't provide measurements to prove it, then it doesn't exist, however the fact that someone has not provided objective proof of something doesn't mean that the conclusion is invalid. It just means that it is a personal subjective observation.
With respect to speaker distortions being much higher than amp ones, this is true, and this has been discussed a lot in the multiway forum. An interesting post (can't remember by whom) in the measurements thread in multiway went some way to explaining why the higher distortions in speakers are mostly irrelevant. The reasoning was that it is the higher order distortions (especially odd harmonics) that are most offensive to human ears, and that it is very difficult if not impossible for transducers to create these high order distortions. An Amplifier however has no trouble at all producing these distortions.
The spice sim I did had THD or around 0.016 with 1V input with the Toshiba VAS transistors that Hugh mentioned. But it was around 0.19% with the BF469's! (BD139's were around 0.018) Now as I said earlier this could be that the model I found for the BF469 is flawed, but if this is really the effect that changing to these transistors (which have been used widely in the past as VAS transistors in amps) then in this application at least, I'd say without some mod to the circuit they are not going to perform optimally.
Now of course I could be completely wrong and the 10X higher distortion may be completely insignificant, however my limited knowledge on the subject leads me to conclude that it will be, and that if you can avoid that, then why not!
Tony.
Hello
I use LtSpice, and simulations are good to give an ideas about an amp, but you can have surprise wen doing listening tests.
I did lot of sim on a amp to get -95db of thd and very low HF harmonics in the thd spectrum, but after some thinkering I've lowered the GNFB, it was giving a bit more thd but did sound much better. So low thd do not always give better sound.
Listening tests are the final test over all the simulations we can do.
Bye
Gaetan
Last edited:
Hi all, an interesting think tank today, my 2cents:
Given that imbalance in the LTP gives rise to predominantly even harmonic distortion and illinearity generally results in odd harmonic distortion, there is an opportunity here at the LTP to royally stuff up THD and still arrive at a disireable outcome, given that we are seeking sonic delights and not necessarily Hi-fi audio. Many seem to actively attempt this.
The critically important feature of any particular implementation would seem to be in limiting the range or number of harmonics so that nasty high ones don't appear at the VAS output. There are plenty of threads and documents floating around the forum to support this design goal. After deciding on the topology and set important DC conditions to stable we are left with gain, bandwidth, frequency stability margins etc. to be set by component selection, and tweaking for want of a better word.
It would be naive to think that generated harmonic structure would not be influenced by reactive components in the circuit, particularly where the gain/bandwidth of the device, not just Ft are being tested. That may be an obvious statement but the result is evident here with the devices used in the audio band, far from the Mhz of video application. If you add in the variable reactance of the transistor depletion layer mentioned earlier, you have a dogs breakfast of dynamic variation too. It would seem prudent to shove that lot as far up the inaudibility scale as possible.
Perhaps then, the very low Cob is not such an extravagence after all.
Given that imbalance in the LTP gives rise to predominantly even harmonic distortion and illinearity generally results in odd harmonic distortion, there is an opportunity here at the LTP to royally stuff up THD and still arrive at a disireable outcome, given that we are seeking sonic delights and not necessarily Hi-fi audio. Many seem to actively attempt this.
The critically important feature of any particular implementation would seem to be in limiting the range or number of harmonics so that nasty high ones don't appear at the VAS output. There are plenty of threads and documents floating around the forum to support this design goal. After deciding on the topology and set important DC conditions to stable we are left with gain, bandwidth, frequency stability margins etc. to be set by component selection, and tweaking for want of a better word.
It would be naive to think that generated harmonic structure would not be influenced by reactive components in the circuit, particularly where the gain/bandwidth of the device, not just Ft are being tested. That may be an obvious statement but the result is evident here with the devices used in the audio band, far from the Mhz of video application. If you add in the variable reactance of the transistor depletion layer mentioned earlier, you have a dogs breakfast of dynamic variation too. It would seem prudent to shove that lot as far up the inaudibility scale as possible.
Perhaps then, the very low Cob is not such an extravagence after all.
Last edited:
I actually agree with you. You shouldn't believe anything you read. You should do what I have done. I have made around 20 amp kits to understand how others have done them and played around changing a few bits. Then I learnt a CAD package and designed dozens of PCBS. I had a play with Spice. I bought a scope, a signal generators or 2, another multimeter, a infrared temperature meter, a cap meter or 2, and a few other things. I then got hundreds of PCBs manufactured, purchased thousands of transistors, resistors and capacitors and I am building a numerous amps with only a few components different.
I don't strictly follow the "scientific method" but hey, I am not at uni anymore.
Now, I will occasionally mention a few things I have found but I don't expect anyone to believe it as gospel.
regards
Good grief, Greg. The prof. will be very busy and skint for some time there. I trust he is thinking about bass or extreme treble distortion at 5% levels. Otherwise in the mids they might be considered mumblers rather than speakers. A relative distortion comparison can be closer to 10:1 from the speakers in this range and possibly harmonically benign.
Let me suggest that there can appear to be more distortion contribution from the amplifier in the midrange and here, according to Klipsch et.al is home and hence our constant struggle to purge it of evil.
Interesting. Was that on this amp? What is required in this circuit to lower GNFB?
I only ask as I do not have good knowledge of amplifier design.
Last edited:
Ok but if speaker distortion was insignificant we could use any speaker and derive conclusions based on that one setup. However it is significant enough that people pay thousands of bucks for different speakers. So how is it valid to not try a number of different speakers/room/source combinations before forming conclusions?
In which case why not scrap the THD specification since it doesn't tell us which harmonics contribute what amount?
Your argument is a simplification in that it only takes into account harmonic distortion but not other kinds like group delay from a speaker. Let's consider intermodulation. Is it possible for this to occur through a combination of amplifier and speaker?
Hello rabbitz
It was about another amp.
But I can use the Babby Aksa as an example, if you raise the value of R9 you will have a lower GNFB, but you would also need to raise R1 to keep the 1:1 ratio of the gnfb resistor and the input Z resistor. But you can also lower the GNFB by lower the value of R8.
R9 and R8 control the amount of voltage who goes into Q2 the negative side of the two LTP input transistor, that Q2 transistor are the thd correction transistor.
So with R9 at 47K ohm you have arround 30 db of gain, which is a good ratio for this type of amp.
Generally it's better to have a low gnfb to have a less compress and more open sound. But not all amps are same and the rule need to be a bit adapt for each amps circuits.
Bye
Gaetan
Last edited:
Hugh;
Of course, you are absolutely correct in your description of the non-linear effects of bjt capacitance in the VAS - I have found that high-voltage bjts (300V and up) seem to be much more linear (hence the use of VIDEO transistors (!)).
However, nowadays I use a Supertex high-voltage depletion-mode MOSFET as the VAS in my designs.
Of course, you are absolutely correct in your description of the non-linear effects of bjt capacitance in the VAS - I have found that high-voltage bjts (300V and up) seem to be much more linear (hence the use of VIDEO transistors (!)).
However, nowadays I use a Supertex high-voltage depletion-mode MOSFET as the VAS in my designs.
Hi Bala,
Fascinating - which one do you use? TN2124 at 240V and 134mA looks good. This is most interesting.... I can see that the lower transconductance of the mosfet is no problem, and will in fact lead to much lower loop gain, which in turn will likely sound better.
Can you share?
Hugh
Fascinating - which one do you use? TN2124 at 240V and 134mA looks good. This is most interesting.... I can see that the lower transconductance of the mosfet is no problem, and will in fact lead to much lower loop gain, which in turn will likely sound better.
Can you share?
Hugh
Last edited:
Hi, Hugh
I beg to differ on this one...
A lower cob allow a lower value for the Cdom since
the non linear part of the total capacitance is lower
with such low Cob components..
As an exemple, lets say that the Cdom must be
ten fold higher that a nominal Cob...
Then , a BJT with a 2pF cob and a 20pF Cdom
will have the same capacitance relative non linearity
than a BJT with 20pF and 200pF Cdom, since in the two
case, if the Cob of the said BJTs shift from 0 to
their nominal values , the two exemples will experience
a total capacitance variation of 10%.
cheers,
wahab
Theory, Wahab.
I have not noticed a marked difference in practice between 100MHz and 200MHz devices, one with 5pF, the other with 1.8pF. I may well be wrong, that's easily correctable, but in practice the differences are slight indeed.
However, the sonic differences are telling, I can say that with authority.
Vive la difference, I say.....
Hugh
I have not noticed a marked difference in practice between 100MHz and 200MHz devices, one with 5pF, the other with 1.8pF. I may well be wrong, that's easily correctable, but in practice the differences are slight indeed.
However, the sonic differences are telling, I can say that with authority.
Vive la difference, I say.....
Hugh
Cob cobbling
The rest is clear enough. You say that the difference between 200+20pF
is effectively same as 20+2pF concerning the ratio of fixed to varying capacitance with Cdom and Cob. Ok, it sounds correct if the same proportional variations in Cob occur in both txs.
Maybe I miss a point also, but the differences considered are more likely these: 100+20pF, 118+2pF. It's 10:1 ratio. This is not the same at all for the comparison of linear proportions. Here there will be benefit with smaller Cob to consider.
Hi Wahab, I assume you are just quoting Hugh in the first paragraph.
The rest is clear enough. You say that the difference between 200+20pF
is effectively same as 20+2pF concerning the ratio of fixed to varying capacitance with Cdom and Cob. Ok, it sounds correct if the same proportional variations in Cob occur in both txs.
Maybe I miss a point also, but the differences considered are more likely these: 100+20pF, 118+2pF. It's 10:1 ratio. This is not the same at all for the comparison of linear proportions. Here there will be benefit with smaller Cob to consider.
Maybe I miss a point also, but the differences considered are more likely these: 100+20pF, 118+2pF. It's 10:1 ratio. This is not the same at all for the comparison of linear proportions. Here there will be benefit with smaller Cob to consider.
Hi , Ian
Right, but the exemple i was quoting is for another purpose..
Since a low Cob BJT will allow a lower Cdom than a high cob
siblings while staying in the same linearity range , we ll have
a higher open loop bandwith and consequently a higher
feedback ratio at high frequencies, the frequencies which are
in need of efficient NFB.
Of course, with the same Cdom , the lower cob BJT
will have a better linearity , all things being equal...
Cheers,
Wahab
Thanks Gaetan
I reduced the gain by 3.4dB on my current AKSA by increasing R8 by 50% which was the maximum recommended by Hugh. To reduce it further I used a 22K/22K dividing network to cut out another 6dB so it matches better with a pre amp with 10dB of gain. I'll need to do a similar thing with the Baby AKSA.
So for the Baby AKSA I would increase R8 to 2K2 to give me around 27dB gain. I thought gain and GNFB were essentially the same thing and a bit confused by decreasing R8 to reduce GNFB where I increased R8 to reduce gain.
Cheers
Peter,
The feedback issues are subtle. Any amp has an open loop gain (OLG) which in fact is the unrestricted gain if the amp is effectively operated without feedback. Obviously, in this form the amp is highly non-linear, but it's an essential parameter if we want to know the details of the feedback loop. You achieve this by putting an infinitely large cap (10,000uF in practice is normally enough) from the feedback node straight to ground, then put in a small, say 1mVp signal, and then measure the peak output of the amp loaded. This might be 30Vp for example. This would then empirically indicate an open loop gain (OLG) of 30/0.001 = 30,000 which is 89.5dB.
We then set the closed loop gain (CLG) by examining the ratios of the feedback network resistors. In the case of say 33K series resistor to output, and 1K to shunt cap, we have a closed loop gain (CLG) of (33 + 1)/1 = 34, or 30.6dB.
Now, armed with OLG an CLG, we can calculate Loop Gain, which is effectively the ratio of OLG to CLG. In this case, using dBs, a log scale, we merely subtract to simulate division, and we have a Loop Gain of (89.5dB - 30.6dB) = 58.9dB. This in fact is sometimes described as 'This amp has 58.9dB of negative feedback', so you get the idea.
Clearly if you change the closed loop gain (the 'gain' of the amp) you will affect the Loop Gain, and this figure is pivotal to stability issues for the amp in an operational environment, with a real load. In fact, it's probably the most vexing issue when designing an amp, since it has to drive a diversity of loads, some of them very nasty.
Hugh
The feedback issues are subtle. Any amp has an open loop gain (OLG) which in fact is the unrestricted gain if the amp is effectively operated without feedback. Obviously, in this form the amp is highly non-linear, but it's an essential parameter if we want to know the details of the feedback loop. You achieve this by putting an infinitely large cap (10,000uF in practice is normally enough) from the feedback node straight to ground, then put in a small, say 1mVp signal, and then measure the peak output of the amp loaded. This might be 30Vp for example. This would then empirically indicate an open loop gain (OLG) of 30/0.001 = 30,000 which is 89.5dB.
We then set the closed loop gain (CLG) by examining the ratios of the feedback network resistors. In the case of say 33K series resistor to output, and 1K to shunt cap, we have a closed loop gain (CLG) of (33 + 1)/1 = 34, or 30.6dB.
Now, armed with OLG an CLG, we can calculate Loop Gain, which is effectively the ratio of OLG to CLG. In this case, using dBs, a log scale, we merely subtract to simulate division, and we have a Loop Gain of (89.5dB - 30.6dB) = 58.9dB. This in fact is sometimes described as 'This amp has 58.9dB of negative feedback', so you get the idea.
Clearly if you change the closed loop gain (the 'gain' of the amp) you will affect the Loop Gain, and this figure is pivotal to stability issues for the amp in an operational environment, with a real load. In fact, it's probably the most vexing issue when designing an amp, since it has to drive a diversity of loads, some of them very nasty.
Hugh
would it bean exaggeration to state that the sound of this particular amp is determined by the 58.9dB of feedback and the compensation that is used with this level of feedback?
If so, then any change in the feedback of 58.9dB will result in this amp sounding different and quite possibly becoming much less stable.