True. I've got some small JW Miller 0.5 Hy ferrite chokes on order to try this filter approach with my HT amp, but if I detect any significant sonic degradation or noise induction due to adding this choke (as I hear with a lot of magnetic components, particularly ferrite ones), I'll just forget about adding the second order input filter, since I'm not willing to add any active stages to my design to do, say, a Sallen & Key type active.
Of course, it has occurred to me that maybe I can 'cheat' and use the common cathode connection of the input differential stage as part of a S&K, on the assumption that the differential voltage between the two input grids is 'always' considerably less than the signal input voltage, but that is not what I would technically consider a 'good design practice' and may result in less than satisfactory results. Might be worth a try, though. It's 'only' DIY
Of course, it has occurred to me that maybe I can 'cheat' and use the common cathode connection of the input differential stage as part of a S&K, on the assumption that the differential voltage between the two input grids is 'always' considerably less than the signal input voltage, but that is not what I would technically consider a 'good design practice' and may result in less than satisfactory results. Might be worth a try, though. It's 'only' DIY
Hassle it is. But we're braking our bones on time-domain reproduction for ages aswell, its not any less hassle. Today it might be even easier to create 80Million programmable pure sinewave generators than fight with all the DACs, amps. Who knows..
There is something interesting in ability to reproduce whole record by storing its spectra. Distortion somewhere in the middle of the record would change whole spectra alot (bit-exact wise). What is our hearing system detecting? Is it detecting time-domain value, or is it detecting spectrum average, or is it detecting phase-spectrum? How many seconds of listening are spoiled by single distortion? If we compare spectrums of records, then its probably completely spoiled, bit by bit. Sure we humans can "forgive", but how many error-seconds?
I only see distortion measures being showed as harmonics that FFT calculates. Any phase deviations will pop up as harmonics, but what we notice quite well is phase information. Changes to phase coherence aren't considered very dramatic, as RC filters haven't been banned yet, but we find monumental efforts going into fighting with jitter in DACs. I wonder if this directly hints that what is audible is phase jitter of amps, and not so much additional harmonics they produce, or linearily fading phase errors, which should be quite a characteristic of acoustics.
Amplification delay depends somewhat on how small a difference is, so deep feedback with lots of compensation seems like candidate for jitter considerations. I wonder, if low NFB amps are avoiding variance of phase delays and that is one reason they sound more natural. Afterall, our ears have no expectations in regards to what harmonics any source ought to produce, they only expect that to be consistent in space and time.
Nonlinearity of course can be taken as phase jitter too, I'm just wondering if its more correct to consider phase distortions instead of just harmonics? That would of course ask for notion of instant frequency and its phase that FFT fails to offer.
thoriated said:
Of course, it has occurred to me that maybe I can 'cheat' and use the common cathode connection of the input differential stage as part of a S&K, on the assumption that the differential voltage between the two input grids is 'always' considerably less than the signal input voltage, but that is not what I would technically consider a 'good design practice' and may result in less than satisfactory results. Might be worth a try, though. It's 'only' DIY
That has been done before! I think I've seen it in the double barreled amplifier, published by Leach in Audio in 1980. But I cannot find my copy back of this one. Maybe someone else can confirm this. The version on Leach's website is a little different and uses just a passive input filter.
Steven
I have previously made this comment in other threads.
Some of you quote others omitting the note about who wrote the
text quoted, which should be automatically inlcuded. Could
you please try to include that piece of info so one doesn't
have to search backwards to figure out who wrote it.
Thanks in advance
Some of you quote others omitting the note about who wrote the
text quoted, which should be automatically inlcuded. Could
you please try to include that piece of info so one doesn't
have to search backwards to figure out who wrote it.
Thanks in advance
First of all, sorry for these questions. I'm an amateur, not like you guys.
What is Lag Compensation?
In LC audio Zapsolute or End millenium design, the RE is about 1k (compared to 22-100ohm for ordinary differential stage). It is about setting gain due to non feedback design, but from Steven's explenation, it is also good for TIM.
What is large enough closed loop bandwidth? How can we make it?
What is Fast amplifier? It is more about the circuit, or more about the selection of transistor?
What is Lag Compensation?
In LC audio Zapsolute or End millenium design, the RE is about 1k (compared to 22-100ohm for ordinary differential stage). It is about setting gain due to non feedback design, but from Steven's explenation, it is also good for TIM.
What is large enough closed loop bandwidth? How can we make it?
What is Fast amplifier? It is more about the circuit, or more about the selection of transistor?
lumanauw said:
The following webpage has a definition of Large and Small Signal Bandwidth: Follow the link to Slew Rate for a definition and example:
http://www.skylondaworks.com/sc_bw.htm
Transistor selection and circuit topology is important to achieve an adequate bandwidth. In general, small signal transistors have higher bandwidths and should be used where appropriate, considering breakdown voltages, peak and average currents, power dissipation, SOA, etc. Surface mount parts usually have smaller parasitics and allow for less PCB layout parasitics. This can improve frequency response, but may not be necessary for audio. A topology that uses less parts, but performs the job, is better. Using lower valued resistors usually increases bandwidth. The trade off can be higher power dissipation and lower efficiency with linear amps. Depending on your music source cd, record, etc., a large signal bandwidth of 30 KHz to 100 KHz with little phase shift is a good start for an amp design. Achieving this and keeping TIM, IM, and THD low enough to not be heard by serious listeners should be good enough.
Depending on your amplifier topology and overall voltage and current gain requirement, you will find that for a large signal bandwidth of 100KHz, the small signal bandwidth will likely be much higher. You could end up with up to 1 MHz or higher, small signal bandwidth.
thoriated said:
Just a note about triangular/sawtooth wave, I did for more then 10 years ago a measurement on a Dynaudio D28AF tweeter, it had actually in some frequency range a triangular/sawtooth wave shape.
If I remember it right it was at quite low frequency, say somwhere within 1 to 5 kHz.
However I made an empiric test and added a 5mm thick plate with an extention of the tweeters own plate/opening, 45 degrees edge, anyhow the triangularity was away and there was only a perfect sine wave in the entire tweeter range from 1 to 20 kHz.
I used just an tone generator so the power rate was also very low.
It's the opposite of lead compensation. It means that the phase of the output of the compensator lags behind the input phase; there is a phase delay.
The Miller capacitor on a VAS is a form of lag compensation. Above the pole frequency the output phase tends towards a 90 deg lag from the input signal. The input signal is usually a current from the diff stage.
lumanauw said:
Miller-effect is generally used for frequency compensation of an amp. Miller-effect is the amplification effect on feedback capacitance. Using a capacitor feedback from the output stage to an input stage or somewhere in-between in an amp with several stages of amplication, can be used to narrow the frequency band or roll-off the frequency response of an amp. The higher the gain, the higher the effective capacitance or feedback. This is used in some opamps. Because the amount of feedback varies with the gain of the transistor(s), it has a self compensating effect on the bandwidth. The beta of a bipolar transistor varies from lot to lot and with temperature. Using the Miller-effect with a feedback capacitor, tends to keep the bandwidth constant with these gain variations.
A decent Microelectronics college textbook will give a better explanation and even some examples of the Miller-effect.
One way to picture phase shifting in an amplifier is with the aid of Bode plots. Looking at the gain magnitude and phase of an amplifier can be very useful. This can be simulated with Spice or measured on the actual circuit using a network analyzer.
You can learn a lot from reading application notes on opamps and power opamps. For example, go to the following webpage and download the PDF for the LM3886:
http://www.national.com/pf/LM/LM3886.html
Scroll down the data sheet and look at the Large Signal Response graph and more specifically the Open Loop Frequency Response graph. You can see how the gain magnitude and phase changes with frequency.
traderbam said:
To add to this:
For stability reasons only one dominant pole in the closed loop is allowed. Unfortunately there are many poles in an amp. By using an extra capacitor across the VAS the pole of the VAS is made dominant and doing so shifting the other poles out of the way. This is the standard technique of making an SS amp stable. It is however a brute force technique IMHO and I doubt this is the optimal way to go.
It is my opinion that the way to go is to design for maximum bandwidth and linearity and compensate as least as possible. There is a school of amp designers that favours constant (but moderate) open loop gain across the entire audio band (up to 20 kHz) but this is difficult to archive.
Cheers
Pjotr said:
I agree with you on all your points.
"By using an extra capacitor across the VAS the pole of the VAS is made dominant and doing so shifting the other poles out of the way."
If I understand you correctly, put another way, the extra compensation capacitor reduces the bandwidth/gain and separates this dominant pole from the others, in frequency. By doing this, the gain crosses zero with adequate phase margin.
I think that many power amp designs have inherent limitations on achieving adequate bandwidth and low enough TIM, IM, and THD distortion, simultaneously. I'm sure many have succeeded, but I see a lot of conventional designs that miss the mark.
A topology that separates the amp functions can make this easier to achieve and I'm sure many such ways exist. Having an amplifier stage for just voltage gain and another stage for current gain without tying the 2 together with a global feedback loop is one such approach. This method can yield a 100 KHz large signal bandwidth without too much distortion.
Sometimes by not defining bandwidth a misunderstanding is created. I think it's important to differentiate between small and large signal bandwidths. For example, a large signal bandwidth of 100 KHz could be defined as: 1 volt peak input and 50 volts peak output from DC to 100 KHz.
lumanauw,
In real basic speak, each pole adds 90deg to phase shift, eventually (45 deg at the pole frequency). It is a sort of smooth phase change. Each pole also causes the voltage gain to reduce by 6dB/oct (halving for every doubling in frequency). If you are beyond the second pole frequency the gain rolls off at 12dB/oct. If there is feedback around the system you, ideally, need the loop phase lag to be <135deg when the loop gain drops below 1.
So you can afford to be at but not beyond the 2nd pole frequency when the loop gain is 1. Because at the 2nd pole f the phase lag will be 90+45=135deg.
Now, an uncompensated amp, say with an op-amp topology but no Miller cap, will have it's 1st pole up in the 100's of kHz and it's 2nd pole not much further in the low MHz. The gain is usually so high that there isn't enough frequency difference between the poles to allow the gain to drop to below 1 at the 2nd pole f. So what people do is introduce a brand new pole at a really low frequency - like 1kHz - so that by the time the 2nd pole is reached (the original 1st pole) the gain is below 1.
A miller cap is a clever way to introduce a low f pole. A simpler way would be to put a capacitor between VAS output and ground. This would have a roughly similar effect to a Miller cap. The Miller arrangement is almost always used because it reduces the amount of current required to charge the cap up and down. The beta of the VAS transistor (if bipolar without emitter resistor) multiplies the effective value of the capacitor, so a much smaller capacitor is needed.
Of course this is over simplified. Real transistors don't have simple poles, especially at high frequencies, so if you plot the phase of a real circuit it won't do nice smooth transistions in multiples of 90deg. It'll do wild things above a few MHz. There are all sorts of invisible parasitic effects and invisible feedback paths around your circuit that make a mess of the simple model. Because of this, it is generally a good idea to make sure your loop gain is below 1 by 1MHz or so.
In real basic speak, each pole adds 90deg to phase shift, eventually (45 deg at the pole frequency). It is a sort of smooth phase change. Each pole also causes the voltage gain to reduce by 6dB/oct (halving for every doubling in frequency). If you are beyond the second pole frequency the gain rolls off at 12dB/oct. If there is feedback around the system you, ideally, need the loop phase lag to be <135deg when the loop gain drops below 1.
So you can afford to be at but not beyond the 2nd pole frequency when the loop gain is 1. Because at the 2nd pole f the phase lag will be 90+45=135deg.
Now, an uncompensated amp, say with an op-amp topology but no Miller cap, will have it's 1st pole up in the 100's of kHz and it's 2nd pole not much further in the low MHz. The gain is usually so high that there isn't enough frequency difference between the poles to allow the gain to drop to below 1 at the 2nd pole f. So what people do is introduce a brand new pole at a really low frequency - like 1kHz - so that by the time the 2nd pole is reached (the original 1st pole) the gain is below 1.
A miller cap is a clever way to introduce a low f pole. A simpler way would be to put a capacitor between VAS output and ground. This would have a roughly similar effect to a Miller cap. The Miller arrangement is almost always used because it reduces the amount of current required to charge the cap up and down. The beta of the VAS transistor (if bipolar without emitter resistor) multiplies the effective value of the capacitor, so a much smaller capacitor is needed.
Of course this is over simplified. Real transistors don't have simple poles, especially at high frequencies, so if you plot the phase of a real circuit it won't do nice smooth transistions in multiples of 90deg. It'll do wild things above a few MHz. There are all sorts of invisible parasitic effects and invisible feedback paths around your circuit that make a mess of the simple model. Because of this, it is generally a good idea to make sure your loop gain is below 1 by 1MHz or so.
mwh-eng said:
No Mark, I was actually not very clear. By making the VAS pole dominant the other poles do not move of course. But the total Pole Zero plot shift. With a given DC gain we have a single point on the plot, the result of all poles and zero’s. And that point moves to a more safe part.
But this is for small signal only. By using a VAS capacitor, the stage driving the VAS needs to be capable of driving enough current to it. Otherwise we have big TIM (slew limiting in the VAS). At large signal this current need can be severe at high frequencies giving higher distortion from the input stage. However it can be tamed by limiting input bandwidth by use of a passive 1st order input low pass filter as suggested by Steven.
With regard to this Steven made also a sometimes-overlooked point: The input stage must be capable of handling the error voltage without giving to much rise of distortion of the input stage.
Cheers
mwh-eng said:
Exactly. Here's a reference that discusses that. http://www.stanford.edu/class/ee214/Handouts/HO9POLESPLIt.pdf
Pjotr,
You may be half right. I imagine in a circuit with a Miller feedback feeding the impedance of the output stage that the phase and gain impact of the output stage poles and zeros on the loop response are changed by the miller feedback. I mean that the VAS and output stage and output load form a feedback system. A little complex to calculate without a simulator.
You may be half right. I imagine in a circuit with a Miller feedback feeding the impedance of the output stage that the phase and gain impact of the output stage poles and zeros on the loop response are changed by the miller feedback. I mean that the VAS and output stage and output load form a feedback system. A little complex to calculate without a simulator.
Unwanted component and PCB layout parasitics must be dealt with one way or another.
I have more experience with switch-mode power supplies than audio amps, but some of the same principles apply to the loop gain. Can have unconditional and conditional stability. I don't like conditional stability.
Using the Miller-effect/cap, from the output to input on one amplifier design, the capacitor needed, became too small (0.5 pf) to be practical in my layout. I moved the feedback point closer to the output to get a higher value of capacitance.
Mostly for the exercise, I'm working on an amplifier design that has a small signal bandwidth greater than 10 MHz. The large signal bandwidth is about 1 MHz. This is with 1 volt peak input and 30 volts peak output. The phase shift at 1 MHz is very small. I can use a linear amp with this much bandwidth for other applications besides audio. If used for audio, I will limit the large signal bandwidth to 100 KHz.
I have more experience with switch-mode power supplies than audio amps, but some of the same principles apply to the loop gain. Can have unconditional and conditional stability. I don't like conditional stability.
Using the Miller-effect/cap, from the output to input on one amplifier design, the capacitor needed, became too small (0.5 pf) to be practical in my layout. I moved the feedback point closer to the output to get a higher value of capacitance.
Mostly for the exercise, I'm working on an amplifier design that has a small signal bandwidth greater than 10 MHz. The large signal bandwidth is about 1 MHz. This is with 1 volt peak input and 30 volts peak output. The phase shift at 1 MHz is very small. I can use a linear amp with this much bandwidth for other applications besides audio. If used for audio, I will limit the large signal bandwidth to 100 KHz.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.