DIY RF Condenser Mics

[Khron]

I have no idea whether much (if any) work on this topic had been done beforehand, but had i been the (only?) one who prompted that research / analysis, at the time?

I do recall having read about the overly-bright K67/K87 capsules being "tamed down" in the electronics. I figured if Neumann themselves "can" do that in their own circuit, and thus it's not "heresy" to alter the response of a given capsule, couldn't a similar effect be achieved with the nearly-ubiquitous Schoeps circuit?

PS: To achieve a high-shelf response, wouldn't one need a resistor in series with the capacitor, across the drain resistor? The cap alone would cause a (monotonous?) high-cut. But i stand to be corrected, of course

In https://groups.yahoo.com/neo/groups/micbuilders/files/Ricardo/ChinaMod%2BU87hybrid/ I analyse adding bits across the drain resistor in detail.  You have to join.

Since then Abbey has provided me with an LTspice model of U87 which is more accurate than what I used with the linear circuit analysis DOS programme I wrote circa late 80's.  This beach bum needs to get off his arse and update ChinaMod+U87hybrid.doc but the findings there are all still valid.  In fact my mod is ideally suited to EQ for AMX8 with the capsule rogs used, where a HF shelf is appropriate.  Only a single cap is required.

I second everything Abbey & Khron says about Henry's EVIL resistors including stuff about Minimum Phase which was of great interest to me in 1980.  I would add that my EQ mod REDUCES NOISE while his increases noise.  If you look carefully at Henry's measurements, you'll see that he actually does have increased noise.

 

[ricardo]

I do recall having read about the overly-bright K67/K87 capsules being "tamed down" in the electronics. I figured if Neumann themselves "can" do that in their own circuit, and thus it's not "heresy" to alter the response of a given capsule, couldn't a similar effect be achieved with the nearly-ubiquitous Schoeps circuit?

ChinaMod+U87hybrid.doc in MicBuilders does exactly this and includes the research that went into the choice of EQ.

Adding a drain capacitor does indeed end up in a "net" high-cut, albeit very slight. Disconnecting the source cap accentuates the high-cut

AMX8 already has this capacitor.  It is C13 1n which matches C9 in the source.  For the HF shelf to make AMX8 match NT1, just increase C13 to 5n6 or even larger ... assuming the pink noise responses are representative bla bla ... I would want to do 'proper anechoic measurements' bla bla ...

I'm moving towardsview that the HF lift in AMX8 being due to RF stuff.  This is cos the noise becomes 'whiter than white'.  I'm too lazy to investigate cos it make my one brain cell hurt.  :-[

In the meantime, I'm listening to the quiet parts of rogs files raised by 40dB.  I hear a couple of quiet 'clonks' which aren't due to the mike so I'll edit them out and redo the noise measurements to see if this makes a difference.

In the meantime, AMX8 does sound slightly 'hissier' than NT1 but at such a low level that it is almo...ost drowned by the '1G' noise.  No idea how  a RF mike has '1G' noise 

 

[Khron]

I hadn't read that document in sooooooome time, but according to the first paragraph, seems like it was indeed me that "prodded" (to some extent, anyway) that research ;D What have i done?.....

Jokes aside though - is the C9 essential to the functioning of the circuit, actually? I mean, apart from the (admittedly outside of audible range) HF boost it does...

 

[rogs]

- is the C9 essential to the functioning of the circuit, actually? I mean, apart from the (admittedly outside of audible range) HF boost it does...

C9 and C13 are part of the time constants formed together with the drain and source resistors R10 and R4.
C9 and C13 are charged every half cycle of the 10MHz oscillator, the voltage being determined by the amplitude of the changing bridge imbalance.
When the FET is turned off during every other half cycle, C9 and C13 start to discharge through R10 and R4.
These time constants determine how fast the circuit responds, and with the values set  one can expect an HF  (-3dB) cutoff frequency of around 30KHz.

In this configuration the FET  - together with R10/C13 and R4/C9 -  is performing 3 tasks simultaneously ---
• Acting as a self biased N channel JFET
•  2 x half wave rectifiers with opposite polarities (an infinite impedance detector)
• An audio frequency phase splitter

 

[ricardo]

I hadn't read that document in sooooooome time, but according to the first paragraph, seems like it was indeed me that "prodded" (to some extent, anyway) that research ;D

I'd forgotten that.

In the meantime, I'm listening to the quiet parts of rogs files raised by 40dB.  I hear a couple of quiet 'clonks' which aren't due to the mike so I'll edit them out and redo the noise measurements to see if this makes a difference.

Done  that.  On the quietest of the quiet sections (the Pink Noise sample), something starts up just before the 'clonks' too.  rogs, have you got Central Heating or a Fridge which starts up 'silently'?

Editing out all this makes minimal difference to the noise.

Hence, NT1 and AMX8 with this capsule are about equivalent noise in practice.  NT1 has a bit more 'whoosh' and AMX8 a bit more hiss but these contributions seem about equal.

If C13 is increased to 5n6 or more, I'd expect response to closely match NT1 and noise 3 or 4 dB better all over.  That's assuming the pink noise measurements are 'representative' of the response of both mikes.  This should be easy enough to check.

What I'd really like to know is

• if it is RF stuff giving the HF lift (and also making the noise 'whiter than white')
• what is causing the '1G' red noise in this RF mike

We can test the first by putting that same capsule in a Schoeps HiZ circuit.

 

[abbey road d enfer]

Adding a drain capacitor does indeed end up in a "net" high-cut, albeit very slight.

It's a matter of value. Since the signal is low-passed only on one half, the ultimate attenuation is 6dB, so it's actually a shelf, rather than a low-pass. I would think this is perhaps better suited to fixinf the sibilance of some capsules than a true low-pass filter.

PS: "Even" the drain-cap-only version is "only" 1.4dB down @ 20kHz (relative to the flat region, at -0.5dB or so), and 0.4dB down @ 10kHz.

4.7k and 1nF results in a low-pass filter tuned at about 35kHz; no wonder it does not produce significant attenuation at 20kHz.

PS(2): Admittedly, i have no clue how the input RF and the demodulation plays into that, but...

Someone, I believe Rogs, mentioned the values chosen in the detector resulted in a 9kHz turnover frequency.

 

[rogs]

4.7k and 1nF results in a low-pass filter tuned at about 35kHz; no wonder it does not produce significant attenuation at 20kHz.
Someone, I believe Rogs, mentioned the values chosen in the detector resulted in a 9kHz turnover frequency.

The original diode based detector had a time constant of 10uS (10k and 1nF) resulting in a turnover at around 16KHz.

This preliminary  version using an FET as an infinite impedance detector used 4k7 resistors to better suit the 'Q' position of the J113 FET.
With the values of C9 and C13 unchanged, that turnover is now c.34KHz.
I'm sure experimenters will be able to change those values to suit their own requirements....

I am having trouble understanding how changing just C13 to 5n6 in this 'balanced' configuration would work?...
Surely that would simply make the value  of the R10 / C13 time constant around 26uS - and change the Fc on the Q2 side of the output to around 6KHz - while leaving the  R4/C9 (Q3) output side at 34KHz?....

 

[abbey road d enfer]

The original diode based detector had a time constant of 10uS (10k and 1nF) resulting in a turnover at around 16KHz.

This preliminary  version using an FET as an infinite impedance detector used 4k7 resistors to better suit the 'Q' position of the J113 FET.

OK, my memory was vague and I didn't want to parse through the whole thread to be more correct.

With the values of C9 and C13 unchanged, that turnover is now c.34KHz.

Ok, these are the values used by Khron in his sim.
So that is essentially flat up to 17kHz.

I am having trouble understanding how changing just C13 to 5n6 in this 'balanced' configuration would work?...

  In that case, the drain output decreases by about 8dB at 13kHz, which is a factor 0.4, so the total differential output at this frequency is 1.4 Vin, which is 3dB below 2. The LF signal is 2Vin. So the response is a shelf that goes from 2 to 1 (0 to -6dB) with a -3dB transition at 13kHz.

 

[rogs]

  In that case, the drain output decreases by about 8dB at 13kHz, which is a factor 0.4, so the total differential output at this frequency is 1.4 Vin, which is 3dB below 2. The LF signal is 2Vin. So the response is a shelf that goes from 2 to 1 (0 to -6dB) with a -3dB transition at 13kHz.

Thanks for explaining that...

 

[ricardo]

If C13 is increased to 5n6 or more, I'd expect response to closely match NT1 and noise 3 or 4 dB better all over.  That's assuming the pink noise measurements are 'representative' of the response of both mikes.  This should be easy enough to check.

Finally resurrected my attempts to become a pseudo LTspice guru.  8)

C13 needs to be 10n for this capsule in AMX8 to match NT1 response.  Then it would have better noise all round.  Of course this holds only for this capsule and assumes the pink noise response  measurements are 'representative' .. bla bla

The shelf introduced is 4.6dB so this will still sound slightly brighter than NT1.  You could tweak this further with R10 but that would increase output level even further when it is already much higher than usual.

At these levels, some attention should be paid to the Schoeps output stage.  Overload is governed by 2 things

• the current taken by the mike on each emitter leg is a voltage across each 6k8 resistor in the preamp.  Taking the maximum sensible current 7mA gives it maximum +ve headroom
• base collector voltage determined by the bias resistors R5 & 6 sets the -ve headroom.  I would make these 150k or even larger

 

[rogs]

Many thanks for the new suggestions ...I shall  measure them in 'real life' as soon as I get a chance...

 

[abbey road d enfer]

Many thanks for the new suggestions ...I shall  measure them in 'real life' as soon as I get a chance...

I've started the simulation but there are too many unknown ATM.
What DC voltage are you measuring at the PNP's collectors and at the BC549 ' collector?
What current through the FET?
What RF voltage at T1 and T2's pri and sec? (I want to determine the turn ratio).
Actually, the tuning of the RF xfmrs seems difficult to figure out since it must be off for proper linear demod but not too much as to offer good sensitivity.
I get to see how demod happens (via the FET's 3/2 characteristic) but my results are probably off by a whole order of magnitude.

 

[rogs]

I've started the simulation but there are too many unknown ATM.
What DC voltage are you measuring at the PNP's collectors and at the BC549 ' collector?
What current through the FET?
What RF voltage at T1 and T2's pri and sec? (I want to determine the turn ratio).
Actually, the tuning of the RF xfmrs seems difficult to figure out since it must be off for proper linear demod but not too much as to offer good sensitivity.
I get to see how demod happens (via the FET's 3/2 characteristic) but my results are probably off by a whole order of magnitude.

• D.C. voltage at Q2 and Q3  (PNP) collectors: 25.3V

• D.C. voltage at Q1 (BC549) collectors: 4.8V

• Using a 10Mohm DMM, the measured 'DC' voltage across R10 and R4 (4k7) resistors (includes RF content of course) depends on the bridge imbalance:

-- With a C12 style capsule (measured as c.90pF), and a 2x47pF in parallel value for C4 , the voltage across R10 and R4 is 4V.

-- With a K.67 style capsule (measured as c.65pF) , and C4 fitted as a 68pF, the voltage across R10 and R4 is 2.5V. 

• I don't have any way of accurately measuring the RF voltages on the  transformers - my best scope lead ( x10  10Mohm 15pF) still loads the transformers significantly.
The turns ratio of the Spectrum 5u3H IF transformers used are included in the highlighted specs attached to this post.

• I'm sure you're right about the tuning settings. The most effective way of tuning both coils I've found so far is the one suggested by Ruud early on the thread.... Tune for maximum  sensitivity, using a tone  applied via a transducer placed against the mic.  Not necessarily a very 'scientific' method I'm afraid, but the only one I've found so far!.....
I did wonder earlier if the actual tuning was 'skewed' from maximum amplitude (post#186)  but decided there would be two optimum settings - one each side of the peak - if that was the case. ...However, so far I've only been able to discover one optimum setting for the tuning in each set up...

 

[rogs]

Next step, figuring out how to do multipattern? ;D

Maybe something along the lines of the attached .....(not tried at all yet!  )

 

[abbey road d enfer]

• D.C. voltage at Q2 and Q3  (PNP) collectors: 25.3V

• D.C. voltage at Q1 (BC549) collectors: 4.8V

• Using a 10Mohm DMM, the measured 'DC' voltage across R10 and R4 (4k7) resistors (includes RF content of course) depends on the bridge imbalance:

-- With a C12 style capsule (measured as c.90pF), and a 2x47pF in parallel value for C4 , the voltage across R10 and R4 is 4V.

-- With a K.67 style capsule (measured as c.65pF) , and C4 fitted as a 68pF, the voltage across R10 and R4 is 2.5V. 

• I don't have any way of accurately measuring the RF voltages on the  transformers - my best scope lead ( x10  10Mohm 15pF) still loads the transformers significantly.
The turns ratio of the Spectrum 5u3H IF transformers used are included in the highlighted specs attached to this post.

• I'm sure you're right about the tuning settings. The most effective way of tuning both coils I've found so far is the one suggested by Ruud early on the thread.... Tune for maximum  sensitivity, using a tone  applied via a transducer placed against the mic.  Not necessarily a very 'scientific' method I'm afraid, but the only one I've found so far!.....
I did wonder earlier if the actual tuning was 'skewed' from maximum amplitude (post#186)  but decided there would be two optimum settings - one each side of the peak - if that was the case. ...However, so far I've only been able to discover one optimum setting for the tuning in each set up...

Thanks for these info. I'll try to make the most of it in my simulation.

 

[ricardo]

I'm sure you're right about the tuning settings. The most effective way of tuning both coils I've found so far is the one suggested by Ruud early on the thread.... Tune for maximum  sensitivity, using a tone  applied via a transducer placed against the mic.  Not necessarily a very 'scientific' method I'm afraid, but the only one I've found so far!.....
I did wonder earlier if the actual tuning was 'skewed' from maximum amplitude (post#186)  but decided there would be two optimum settings - one each side of the peak - if that was the case. ...However, so far I've only been able to discover one optimum setting for the tuning in each set up...

I suspect the capsule and C4 are just acting as a potential divider for the RF signal.  That would explain the single twiddle peak.  ie AM modulation is NOT due to the LC frequency changing.

If so, T2 is unnecessary.  It ups the voltage but this is already too high for comfort.  It also raises the impedance which raises the 'noise resistance'.  Removing T2 would also remove the noise contribution of the transformer and its associated bits.

We would need to do another noise & sensitivity test like #196 & #219 to confirm if the 'equivalent noise spl' is better.  It may be that T2 is what's introducing the HF lift on signal & noise but that might be wishful thinking 

T1's centre tap is not used and the bottom of its secondary is earthed.  That means the capsule can be earthed too which is often convenient.  If C4 is variable it can be set for max sens.  dC in the capsule is most effective when C4 & the capsule are the same.

So alignment is to twiddle C4 for max sens, then T1 for the same thing.

Du.uh!  This is exactly GG Baxandall's procedure & his understanding of the situation

If T1 secondary CT isn't used, that opens up the possibility of using his phase sensitive detector.  Using 2 FETs instead of his 2 BJTs would simplify driving it.

 

[abbey road d enfer]

I suspect the capsule and C4 are just acting as a potential divider for the RF signal. 

They do act as a potential divider between an in-phase and an out-of-phase signal, so the basic unbalance determines the RF base level. perfect balance would mean full-wave rectification of the audio signal.

That would explain the single twiddle peak.  ie AM modulation is NOT due to the LC frequency changing.

My simulation (with all its imperfections, approximations and unknowns) shows that both the tuning AND the balance are changed. See Attachment, which shows voltage across C3 (47pF, C8 on the original schemo). Max gain here is about 20dB, which would be about 30vrms (estimating the output of the oscillator at 3Vrms (collector voltage 4.8V). But it looks like the whole circuit shows a maximum significantly lower than 10MHz; actually, at 10MHz, the RF signal is attenuated.

  dC in the capsule is most effective when C4 & the capsule are the same.

It may be but it's not linear operation when it comes to demod.

This is exactly GG Baxandall's procedure & his understanding of the situation

This seems to contradict posts #147 & 148.

 

[ricardo]

They do act as a potential divider between an in-phase and an out-of-phase signal, so the basic unbalance determines the RF base level. perfect balance would mean full-wave rectification of the audio signal.

Yes.  At perfect balance, it is a DSBSC signal which needs Baxandall's Phase Sensitive Detector to demod w/o distortion

But we don't twiddle C4 to balance the bridge in AMX8.  I think we have a much simpler situation where 'RF leakage from the wonky bridge' is amenable to simple 'diode' rectification.  It's not the LC that needs to be 'detuned' but the bridge.

My proposal maximises this and uses the full RF voltage, 'halved' by C4 & the capsule acting as a potential divider.  As we are not using a bridge & DSBSC, we can twiddle C4 for max audio output.  Gotta be careful as FET bias changes too.

There are problems.  The 'bias' for the FET is the rectified RF voltage.  If RF is 10Vrms, this is 7V across both  R4 & R10

In AMX8, we could twiddle C4 to get a desired bias across R4 & R10 from 'rectified RF leakage' from the slightly wonky bridge

 

[Gerard]

In AMX8, we could twiddle C4 to get a desired bias across R4 & R10 from 'rectified RF leakage' from the slightly wonky bridge

Not sure I quite understand what you are saying; I'm just a dumb 'mechanical' engineer. What is the "desired bias" value. Could you please elaborate a bit.

 

[abbey road d enfer]

But we don't twiddle C4 to balance the bridge in AMX8.  I think we have a much simpler situation where 'RF leakage from the wonky bridge' is amenable to simple 'diode' rectification.  It's not the LC that needs to be 'detuned' but the bridge.

I agree totally. It calls into question the need for a bridge, since it needs to be very out of balance for proper demod to happen.

My proposal maximises this and uses the full RF voltage, 'halved' by C4 & the capsule acting as a potential divider.  As we are not using a bridge & DSBSC, we can twiddle C4 for max audio output.

I agree it makes more sense and makes the design simpler to build and to tune. ATM the interaction between the two tanks make the analysis quite difficult.

The 'bias' for the FET is the rectified RF voltage.  If RF is 10Vrms, this is 7V across both  R4 & R10

Check attachment. That's actually the source voltage vs. capsule capacitance. I have chosen a significant unbalance (fixed cap 68pF, capsule 78 to 86p); as a result, the actual RF voltage at the FET's gate is from 2Vpp to 4Vpp. The resulting source DC voltage varies from 1.8 to 2.7V. That's 0.9V, which would correspond to about 300mV rms or 600mVrms for the full diff output. That's not too different an order of magnitude than the same capacitance variation applied to a 1G resistor with 70V bias, 560mVpp IINM.