Jean Clochet
Well-known member
I've been working on a low noise, low distortion mic-amp design and thought it might be of interest to one or two people.
Warning: It's a long post so feel free to ignore if time/patience/life's too short!
I've attached a PDF of the input stage. The complete mic pre schematic is a large PDF (Maybe I'll post it later) so I've just shown the basic idea for this section which I thought was quite interesting. I added the signal In/Out connections only for illustration in this post but of course, there are the typical input bias & protection components etc. on the real thing.
The gain btw is split between this 1st stage and the 2nd stage (1 X dual pot controlling both gains simultaneously) with the 1st stage holding about 44dB gain and the 2nd about 22dB. There's also an output driver with level control which has 10dB in hand.
A lot of the folks here will be able to see what's going on with just a quick glance at the PDF picture but, for anyone that wants a bit of a description, I wrote something up.
So:
The parallel 2N4403's (Q1-3 & Q4-6) and respective BC550's (Q7& Q8) are connected as complimentary-feedback-pairs (CFP) for the purpose of reducing distortion. However, sometimes there is a compromise decision to be made with a CFP when adopting the usual method of supplying the bias current for the input devices via a resistor (See the project here: http://sound.westhost.com/project66.htm for an example of this). The trade off is one of finding the right balance between noise and distortion reduction:- The resistor value that ensures the appropriate current for the input devices (for the lowest noise at the design-centre input impedance) might also lower the gain available for linearisation via the feedback action of the CFP.
Here, the resistor is eliminated entirely and the 2N4403 collectors are supplied with bias current via constant current sources (bottom of picture). The current here is 3mA for approx. 1mA per input device. Of course, the CCS's need to be quiet ones.
Anyway, we now have maximum feedback action happening for reducing distortions but we also get to pick the optimum current at which the input devices operate. I'd suggest that it's important if you are modeling this in Spice that you model the actual CCS that you'd plan on using rather than an ideal one. Otherwise, you may not pick up on how much or how little noise the CCS is adding to the signal.
The other aspect which I thought was interesting with this input stage addreses Common-Mode-Rejection.
With variable-gain diff-amp stages such as this, CMMR is generally fine at high gains (smaller emitter to emitter resistance values) but degrades at low gains when the emitter to emitter R value is quite high. This is due to early effect of the input transistors.
Here, a cascode is used (the BC560's - Q9 & Q10) to mostly eliminate these effects by keeping the collector/emitter voltage of the input transistors constant - which goes some way, but not all the way, to keeping CMMR constant for all conditions.
The cascode is the enhanced type, as explained by Prof. Malcolm Hawksford (I'll look for the paper and post a link), but, instead of the usual zener or string of LED's to establish the cascode stand-off voltage, a resistor is used here (the 10K R's - R7 & R8) which is generally a bit quieter than if using a diode instead.
The 10K cascode bias resistors are supplied with about .6mA of current via another CCS so, basic maths ohms's law tells us that there's about 6V of cascode bias.
CMMR at all gains is now pretty darn good (mostly dependant on matching) and also quite a bit better at higher frequencies regardless. Tests show that THD is also further reduced with the cascode connection and is independant of frequency (for audio purposes anyway!). The small caps (15pF) in parallel with the 10K cascode bias R's look a bit silly but they do ensure stability.
Since the current to the emitters of the 2N4403's is supplied via resistors (R1 & R2) directly from the positive rail rather than more CCS's (this is for noise reasons) as well as the fact that the 825R load resistors (R3 & R4) are connected directly to the negative rail, PSSR is not exactly stellar for a low noise pre-amp. This is true of all stages such as this and isn't peculiar to this design.
In the version I'm working on, I have the main incoming voltage LM317/LM337 regulated at +/-24V and these then feed fairly simple capacitor multipliers for each channel which drops the supply down to +/- 22V. They ensure the rails are quiet enough for low noise performance.
Additionally, individual cap. multipliers per channel would also give good channel to channel crosstalk measuements.
I'm happy so far with how the complete mic amp is coming along but I'm still learning as I go...
More later maybe but I think that's enough blurbage for now,
cheers.
Edit: Q7/Q8 connections corrected on PDF.
Warning: It's a long post so feel free to ignore if time/patience/life's too short!
I've attached a PDF of the input stage. The complete mic pre schematic is a large PDF (Maybe I'll post it later) so I've just shown the basic idea for this section which I thought was quite interesting. I added the signal In/Out connections only for illustration in this post but of course, there are the typical input bias & protection components etc. on the real thing.
The gain btw is split between this 1st stage and the 2nd stage (1 X dual pot controlling both gains simultaneously) with the 1st stage holding about 44dB gain and the 2nd about 22dB. There's also an output driver with level control which has 10dB in hand.
A lot of the folks here will be able to see what's going on with just a quick glance at the PDF picture but, for anyone that wants a bit of a description, I wrote something up.
So:
The parallel 2N4403's (Q1-3 & Q4-6) and respective BC550's (Q7& Q8) are connected as complimentary-feedback-pairs (CFP) for the purpose of reducing distortion. However, sometimes there is a compromise decision to be made with a CFP when adopting the usual method of supplying the bias current for the input devices via a resistor (See the project here: http://sound.westhost.com/project66.htm for an example of this). The trade off is one of finding the right balance between noise and distortion reduction:- The resistor value that ensures the appropriate current for the input devices (for the lowest noise at the design-centre input impedance) might also lower the gain available for linearisation via the feedback action of the CFP.
Here, the resistor is eliminated entirely and the 2N4403 collectors are supplied with bias current via constant current sources (bottom of picture). The current here is 3mA for approx. 1mA per input device. Of course, the CCS's need to be quiet ones.
Anyway, we now have maximum feedback action happening for reducing distortions but we also get to pick the optimum current at which the input devices operate. I'd suggest that it's important if you are modeling this in Spice that you model the actual CCS that you'd plan on using rather than an ideal one. Otherwise, you may not pick up on how much or how little noise the CCS is adding to the signal.
The other aspect which I thought was interesting with this input stage addreses Common-Mode-Rejection.
With variable-gain diff-amp stages such as this, CMMR is generally fine at high gains (smaller emitter to emitter resistance values) but degrades at low gains when the emitter to emitter R value is quite high. This is due to early effect of the input transistors.
Here, a cascode is used (the BC560's - Q9 & Q10) to mostly eliminate these effects by keeping the collector/emitter voltage of the input transistors constant - which goes some way, but not all the way, to keeping CMMR constant for all conditions.
The cascode is the enhanced type, as explained by Prof. Malcolm Hawksford (I'll look for the paper and post a link), but, instead of the usual zener or string of LED's to establish the cascode stand-off voltage, a resistor is used here (the 10K R's - R7 & R8) which is generally a bit quieter than if using a diode instead.
The 10K cascode bias resistors are supplied with about .6mA of current via another CCS so, basic maths ohms's law tells us that there's about 6V of cascode bias.
CMMR at all gains is now pretty darn good (mostly dependant on matching) and also quite a bit better at higher frequencies regardless. Tests show that THD is also further reduced with the cascode connection and is independant of frequency (for audio purposes anyway!). The small caps (15pF) in parallel with the 10K cascode bias R's look a bit silly but they do ensure stability.
Since the current to the emitters of the 2N4403's is supplied via resistors (R1 & R2) directly from the positive rail rather than more CCS's (this is for noise reasons) as well as the fact that the 825R load resistors (R3 & R4) are connected directly to the negative rail, PSSR is not exactly stellar for a low noise pre-amp. This is true of all stages such as this and isn't peculiar to this design.
In the version I'm working on, I have the main incoming voltage LM317/LM337 regulated at +/-24V and these then feed fairly simple capacitor multipliers for each channel which drops the supply down to +/- 22V. They ensure the rails are quiet enough for low noise performance.
Additionally, individual cap. multipliers per channel would also give good channel to channel crosstalk measuements.
I'm happy so far with how the complete mic amp is coming along but I'm still learning as I go...
More later maybe but I think that's enough blurbage for now,
cheers.
Edit: Q7/Q8 connections corrected on PDF.
