New Transformerless Mic PreAmp

[Samuel Groner]

Hi

I implemented design B from the "Mic Preamp Schematic Collection" thread (www.groupdiy.com/index.php?topic=20038) as I was interested in its performance and because I got asked for a layout from different people.

Here are the important files:
Schematic: monte_generoso_r1.pdf
Layout: monte_generoso_r1_PCB.pdf
Gain switch resistor values: monte_generoso_r1_gain_switch.pdf

The topology used allows complete freedom from electrolytic capacitors and DC servos, at the cost of not perfectly zero output offset. It turned out that DC precision was better than expected--the trimmer was needed for less than 100 uV input related offset correction. It drifts a bit but the output seems to stay within 10 mV offset all the time, which is perfectly fine I'd say.

Gain is set with a 24 position switch from 7.5 dB to 65 dB in 2.5 dB steps. Polarity is switched at the output with a relays.

And now a few measurements:
Normalised frequency response: monte_generoso_r1_freq.pdf
THD+N: monte_generoso_r1_thdn.pdf
CMRR: monte_generoso_r1_cmrr.pdf

Note that the frequency response is almost perfectly equal for all gain settings. Only at the highest frequency shown there is some indication that the first stage runs on somewhat reduced loop gain, although the -3 dB bandwidth of the first stage is still above 1 MHz. The bass response is very pretty--a direct result of having only one AC coupling capacitor pair in the signal path.

The THD+N plots show good performance as well. The residual consists mostly of noise, except for the highest gain where a faint of third harmonic starts to become visible above 5 kHz. If you want to compare this data to other preamps please note that the measurements were done at high bandwidth, somewhat higher-than-usual maximum gain (65 dB instead of 60 dB) and on the bench without any shielding. This will easily increase the THD+N reading by a factor of 4 or more at the maximum gain in contrast to other published data.

Slew rate is typically 110 V/us and EIN has been measured at about -129 dBu for a 150 Ohm source (20 kHz bandwidth, 65 dB gain) and -135 dBu with shorted input.

CMRR at low frequencies is the only thing I'm not absolutely happy with though it surely is enough for real world use and (way) better than many other preamps. The reason for the decreasing CMRR towards the lower range is mismatch in the input coupling capacitors--I didn't realise that this is that critical up to almost the upper end of the audio range. This can be figthed by tightly matching the capacitors C105 and C106 (e.g. by the addition of small compensation capacitors in order to avoid the need for buying a large batch of expensive 33 uF film capacitors). If I'd do another revision (not to be expected soon) I think I'd use the common-mode bootstrap method invented by Bill Whitlock to avoide the matching need.

On the other hand the input RFI filter is very effective indeed (at least as far as I can tell). Above the shown range CMRR very rapidly increases. The onset of this is just visible at 1 MHz. In addition to this I noticed that common-mode signals show very low waveform distortion even at the highest frequencies (not something I usually observe with transformerless designs). That's a result of having high slew-rate amplifiers and will account for low intermodulation distortion with spurious HF products.

Overall very pleasing results, I'd say. I was not able to give it a detailed listen yet (this will--for my convenience--have to wait until it got a case), but I would be surprised if this preamp wouldn't be very clean and transparent.

[Edit--some pictures: monte_generoso.html.]

Samuel

 

[BYacey]

Looks interesting Samuel. Depending on how much out the coupling caps are, perhaps a trimmer cap in parallel with the lower value 33mfd would enable you to balance the cmrr.

Another option would be to put a 100 ohm trimmer resistor in parallel with R117 , and change the value of R117 to 22 ohms. This enables you to trim the CMRR at the highest gain settings.

To Trim the CMRR at low gain settings, open the connection between R119 and R120, and connect a trimmer resistor, perhaps 200 ohm between them with the wiper going to the Vcc source. These are what Valley implemented on the Transamp circuit to enable final trim on the CMRR.

Disregard my quote on the bottom; the part number happens to fall pretty close to the numbers on your schematic!! :grin:

 

[Samuel Groner]

Thanks for your interest and suggestions!

These are what Valley implemented on the Transamp circuit to enable final trim on the CMRR.

You can't compare the Transamp with this design, it's a different topology and at a different level of performance. Basically it is not possible to resistively trim a CMRR error due to capacitor mismatch without introducing another CMRR error elsewhere (e.g. at a different frequency or with different source impedances) as the error is frequency dependent and contains phase shift. The basic CMRR capability of the circuit is absolutely excellent with the four provided trimmers (probably well in excess of 120 dB at medium gain due to the excellent matching of the input transistors and collector loads) and needs no further tweak. See below for a more detailed answer.

Depending on how much out the coupling caps are, perhaps a trimmer cap in parallel with the lower value 33 uF would enable you to balance the CMRR.

The mismatch is in the nF region where it's pretty hard to find a suitable trimmer.

Another option would be to put a 100 ohm trimmer resistor in parallel with R117 , and change the value of R117 to 22 ohms.

This won't work here. Note that in the Transamp there are some resistors (R7, R11 and R12 in the schematic I have at hand) to ground which allow this trim to be effective. In my design they are just in series with the high impedance input and have only very minor effect on CMRR. In addition to this, increasing the resistance here will add thermal noise.

To Trim the CMRR at low gain settings, open the connection between R119 and R120, and connect a trimmer resistor, perhaps 200 ohm between them with the wiper going to the Vcc source.

As noted above, the input stage is allready very well balanced and trimming here will have very little effect (besides having the danger of introducing a serious drift term which might be disastrous to DC precision at high gains).

It's really the fact that the input coupling capacitors show some mismatch and that R109/R111 need to have low DC resistance, and there is no easy way around that (as far as I know...).

Samuel

 

[BYacey]

[quote author="Samuel Groner"]
It's really the fact that the input coupling capacitors show some mismatch and that R109/R111 need to have low DC resistance, and there is no easy way around that (as far as I know...).

Samuel[/quote]
Constant current source ala Valley again. This offers a low DC resistance while maintaining a high AC impedance, an elegant solution.

By trimming on the front end in the schematic you provided a link to, this is just correcting the DC offset that is being created due to imbalance in the gain stages. I would think this would further deteriorate the CMRR rather than getting the gain stages to balance with respect to each other. I don't see a whole lot of difference in the front end of the two circuits (up to the bases) except for the addition of the DC bias. R117 and R118 do little except provide a small value that can be juggled to trim High gain CMRR. Some cheaper preamps don't even provide them.

 

[BYacey]

In looking at your circuit closer, it's not that different from the Valley as you might think. Parallel input transistors driving an opamp stage with negative feedback brought back to the emitters. The only real differences I find is the DC coupling and the more elaborate 2nd opamp stages, and the collector resistors instead of using a constant current source.

 

[Samuel Groner]

Constant current source ala Valley again. This offers a low DC resistance while maintaining a high AC impedance, an elegant solution.

Look again. VP does use current limiters connected to the emitters not to the bases. These current limiters are not needed here because I do DC couple to the opamps which allows to set DC conditions that way (that may look as a small difference at first but a small change goes a long way here). Adding current sources to the emitters would degrade noise figure at low gains and reduce CMRR (though only little considering their hopefully high impedance). And replacing R109/R111 with current limiters as suggested by you really won't work at all (base currents need a resistive path as they are inherently ill-defined and drifty).

And the collector resistors instead of using a constant current source.

As above--the current limiter in the VP schematic goes to the emitter, not collector. They use resistors for the collector load as I do.

By trimming on the front end in the schematic you provided a link to, this is just correcting the DC offset that is being created due to imbalance in the gain stages.

Sure that's the idea of trimmer R115 (as noted on the schematic..?). CMRR is trimmed by R134, R140, C117 and C121 (which are the trimmers I referred to in the previous post).

I would think this would further deteriorate the CMRR rather than getting the gain stages to balance with respect to each other.

No, this does not detoriate CMRR in any significant way. R116 is shunted by R110 and hence input impedance balance is well maintained. Neither does the position of R115 affect gain. The gain stages are well in balance, as inherently with the instrumentation topology (common mode gain is unity for both halves of the first stage as there is no shunt impedance (apart from parasitics) from the emitters to ground) and the precision parts used here. And note that the corrective voltage set by R115 is in the uV (!) range--a good indication that the second stage has truly impressive balance allready.

R117 and R118 do little except provide a small value that can be juggled to trim high gain CMRR.

Again, without having shunt resistors to ground after R117 and R118 they won't trim CMRR in any significant way, sorry. They don't change gain, so how should they affect CMRR? The only reason for having these resistors is to limit current to the protection diodes D101-D104.

Samuel

 

[JohnRoberts]

I haven't been following this closely, but my sense is that the topologies are more similar than different.

The HF CMRR trim ala Buff was probably working into the circuits input capacitcance since there were multiple medium power transistors in parallel to get low noise back in those days, so input C could be a factor.

My preference is to avoid input caps if possible but the sparate LF and then HF trim was the bee's knees back in the day. Modern devices may not benefit as much (I didn't trim either for high volume production using newer, single, low noise devices).

JR

 

[Samuel Groner]

I haven't been following this closely, but my sense is that the topologies are more similar than different.

Well, there are several differences I'd say:
* the quiescent current of the input pair is set by emitter current limiters in one topology while the other uses collector current limiters (or however you like to call this resistor with fixed voltage across it)
* the Transamp does AC coupling to the opamp while the other uses DC coupling
* one design uses one opamp to follow the collector outputs while the other uses two
* the Valley People design provides CMRR entirely in the first stage while the other transfers common mode voltage to the (optional) second stage

Of course I can understand why you and Bill see several similarities. Both provide a differential base input, gain is set by one resistor and feedback goes to the emitters of the input pair. But that all is common to any current feedback instrumentation topology (there are more than just these two)--for me it's like saying that TL071 and OPA627 are similar: both have FET inputs, single-ended output and voltage-feedback.

The HF CMRR trim ala Buff was probably working into the circuits input capacitcance since there were multiple medium power transistors in parallel to get low noise back in those days, so input C could be a factor.

It's not a high-frequency but rather a high-gain trim. It works by adding a voltage divider (22 ohm/100k ohm) before the input.

Samuel

 

[JohnRoberts]

[quote author="Samuel Groner"]

It's not a high-frequency but rather a high-gain trim. It works by adding a voltage divider (22 ohm/100k ohm) before the input.

Samuel[/quote]

My recollection is a little fuzzy since it's been decades since I had one of those (transamps) on the bench. I never felt that even with the two input trims was it ever perfectly balanced. The output differential trim was pretty stable, but two input trims would interact. Quite good when properly adjusted (to procedure) but I have long been inclined to eliminate trims from circuits by prudent design, thus my preference for no input caps (when possible).

By similar I do not mean same. I have designed multiple circuits using discrete devices with opamps wrapped around them. They were all different, but shared sundry common properties that made them similar IMO.

YMMV

JR

 

[bcarso]

I like it, particularly the elimination of the big gain switch C and servos.

In the fairly short time I've been with the forum, it has been a pleasure to observe Samuel's growth as a designer, of which this is further evidence. A style is emerging---a certain fastidiousness, attention to detail.

The relay choice looks to be a good one:

http://www.omron.com/ecb/products/pry/111/g6h.html

Relays and other electromechanical switches always worry me though. Omrons are surely among the best, but even they can develop dry switching problems. Certainly, positioning this one at the end of the amplification chain is the best place for it.

It got me to wondering if anyone has ever produced a circuit with an electromechanical relay contact enclosed in a local feedback loop, that looks at the voltage across the contacts and drives a signal through them until the adsorbed contaminant or corrosion is blasted away. I have seen some circuits in which feedback is applied around the relay contacts, but never something that could be treated as just another relay contact, to be inserted wherever without concern except for the voltage limit.

Of course there's always mercury-wetted reeds, until the dutifully ecology-minded manage to ban them altogether (except for exemptions for war toys, probably).

John R and I talked about strategies we've both used for dealing with the distortions attendant on solid-state switching, out of the thread where I asked about peoples' experience with the overload behavior of CMOS bilateral switches. BTW that huge switched filter bank was completed, and aside from a suboptimal board layout entailing a lot of sensitivity to magnetic fields, works like a charm. The functions still should have been realized with a litle DSP engine I think, and that extra 6 dB or so of range was painful for all of those protection diodes that are needed, but aside from that it works fine.

 

[Rossi]

I'm impressed with Samuel's work as well :thumb: :thumb: :thumb:

I recently had a true systems P2 analog in my studio for review purposes. Its bill of materials is quite similar (2x MAT02, 2x OPA2604 per channel) but different topology, as far as I can tell. Would be interesting to see which of the two would win the competition :grin:

I have a more basic question: would it be possible to substitute the paralleled MAT02 transistors by single super low Rbb types (e.g. 2SC3329 or 2SC2546) Or do these paralleled MAT02s have properties that cannot be had from single transistors?

 

[bcarso]

[quote author="Rossi"]I have a more basic question: would it be possible to substitute the paralleled MAT02 transistors by single super low Rbb types (e.g. 2SC3329 or 2SC2546) Or do these paralleled MAT02s have properties that cannot be had from single transistors?[/quote]

Yes---a high degree of Vbe matching and thermal coupling. With sufficient patience you might be able to find a matched pair of low Rbb parts, but then you would still have the problem of their thermal coupling being no better than that from pressing them together, maybe embedding them in a thermal conductor as well.


EDIT: spelling

 

[Rossi]

I suppose I was a bit unclear. I meant substituting both halves of the MAT02 by just one super low Rbb transistor. Assuming that paralleling the two MAT02 halves was just for lowering Rbb.

 

[bcarso]

You may already note the way the transistors are connected: A and B of a given pair are between the two halves of the circuit. Q101A is paralleled with Q102A, Q101B with Q102B. Hence the thermal coupling and matching is preserved.

Sure, if you can match very closely two low Rbb devices, one for each half could be substituted. But there is still the thermal coupling issue. And I doubt you could easily get that good a match, enough for Samuel's coupling-cap-free and servo-free operation.

Actually, it's interesting that unless the two samples of MAT02 pairs are themselves fairly well matched, there will be one of each in each paralleled pair that tends to hog the current a bit. But since the noise is dominated by Rbb' (I deduce it's about 36 ohms, backed out from AD's datasheet curves at 1mA Ic) it won't make much difference.

It is also interesting that AD mentions bulk resistance, but doesn't give a spec for Rbb'.

 

[Rossi]

[quote author="bcarso"]You may already note the way the transistors are connected: A and B of a given pair are between the two halves of the circuit. Q101A is paralleled with Q102A, Q101B with Q102B. Hence the thermal coupling and matching is preserved.[/quote]

D'oh! Now I see! I hadn't noticed that the two halves are used for both legs. Now it all makes sense.

 

[bcarso]

[quote author="Rossi"]
Now it all makes sense.[/quote]
:grin: :thumb:

 

[StephenGiles]

I've just looked in my SSM chips box and see a PMI MAT04, so presumably this will be OK for 2 channels?

 

[Samuel Groner]

Thanks for the further contributions and your interest!

In the fairly short time I've been with the forum, it has been a pleasure to observe Samuel's growth as a designer, of which this is further evidence.

Indeed there has been some progress--four years ago I didn't know which end of the diode was which... And Brad surely had some major impact on this (not the trick about the diode though :wink: ). :thumb:

I recently had a true systems P2 analog in my studio for review purposes.

Any pictures? Would be cool.

Would it be possible to substitute the paralleled MAT02 transistors by single super low Rbb types?

As Brad said low drift and good initial matching is very important here--every bit of offset gets amplified almost 2000x in the highest gain setting. uVs get mVs... In addition to this, at least the 2SC3329 got quite a bit lower hFE than the parts suggested (unless you are able the source the BL grade), which will increase current noise and 1/f noise (due to the relatively small input coupling capacitors). But that's just a side note which will not matter in practice.

I've just looked in my SSM chips box and see a PMI MAT04, so presumably this will be OK for 2 channels?

Unfortunately a single MAT04 is not equal to a pair of MAT03s. The noise is quite a bit higher. You might be able to build a single channel though by paralleling all four transistors--this will more or less give similar noise performance as a pair of MAT03s. Sharing a MAT04 amongst two channels would anyway be rather difficult as good crosstalk, sufficient stability (at the highest gain the first stage has a gain-bandwidth-product of >1 GHz!) and convenient physical placement will not be easily achieved.

Samuel

 

[StephenGiles]

Many thanks Samuel.

 

[Rossi]

[quote author="Samuel Groner"]

I recently had a true systems P2 analog in my studio for review purposes.

Any pictures? Would be cool.[/quote]

I took a few pictures, but since I handed them in to be published with my article in one of the next issues of S&R, I shouldn't post them on the web. I could send you a pic or two by private mail, though.

What I found interesting was that the enclosure was very deep, family pizza box style. The boards would have fit into a much smaller enclosure, there was a lot of space for separation between preamps and PSU section. If I were to design a transformerless preamp I'd make sure the input capacitors were as far from the mains transformer as possible - as was the case here - and I'd also make sure both capacitors get equal exposure. Which is rarely the case. In most designs one capacitor actually shields the other, so all the hum is in one input leg. Doesn't matter much if there is a lot of space between the PSU and the actual preamp board, but in a more crowded enclosure, there would be some benefit to equidistant cap placement, I think. I once made experiments with a cheapo transformerless preamp. There was noticeably more hum pick-up in the right channel which could be made to disappear by shielding the input lytics.

As Brad said low drift and good initial matching is very important here--every bit of offset gets amplified almost 2000x in the highest gain setting. uVs get mVs... In addition to this, at least the 2SC3329 got quite a bit lower hFE than the parts suggested (unless you are able the source the BL grade), which will increase current noise and 1/f noise (due to the relatively small input coupling capacitors). But that's just a side note which will not matter in practice.

I see. Not a solution in this case, but for other projects, you might consider the 2SC2546 from Hitachi. They don't give a Rbb figure, but it must be very low. In my ribbon booster experiments it was about as low noise as the 2SC3329. The 2SC2546 parts I got from Reichelt were very high hFE, I think in the 700s.

Btw, did you consider the THAT transistor arrays? They offer 4 matched tranistors (PNP, NPN or mixed gender) in one IC-style box. Rbb is spected as 30 ohms.