Thank you for your quite reasonable, and interesting, question, and your acknowledgement of the answer.
There was no inconvenience. I welcome probing questions- they can lead to unexpected insights.
In actuality, this was one of the areas that I "agonized" over when designing the calibration circuit and the associated transimpedance cell.
Incidentally, in case that it's not obvious, the calibration feedback opamp also is a significant contributor to the potential problem, hence the use of a resistive attenuator at its output. This helps to reduce both the distortion and the 1/f noise contribution.
I suspect that I will ultimately change the input network slightly to 1uf/100k/1meg. The rest of the 1uF caps are X7Rs and not suitable, and 1uF C0Gs are not, apparently, available, so the option was a PP or PPS device.
That was not an "optimal" choice as far as availability, size and stock was concerned when I first looked, but I revisited the issue in response to your question, and a suitable PP cap is indeed available at a reasonable cost and availability, and with a suitably small footprint. That enables a further several tenths of a dB in A-weighted S/N ratio to be realized.
There was no inconvenience. I welcome probing questions- they can lead to unexpected insights.
In actuality, this was one of the areas that I "agonized" over when designing the calibration circuit and the associated transimpedance cell.
Incidentally, in case that it's not obvious, the calibration feedback opamp also is a significant contributor to the potential problem, hence the use of a resistive attenuator at its output. This helps to reduce both the distortion and the 1/f noise contribution.
I suspect that I will ultimately change the input network slightly to 1uf/100k/1meg. The rest of the 1uF caps are X7Rs and not suitable, and 1uF C0Gs are not, apparently, available, so the option was a PP or PPS device.
That was not an "optimal" choice as far as availability, size and stock was concerned when I first looked, but I revisited the issue in response to your question, and a suitable PP cap is indeed available at a reasonable cost and availability, and with a suitably small footprint. That enables a further several tenths of a dB in A-weighted S/N ratio to be realized.
Last edited:
I thought that the main purpose of that circuit was to always keep the total Id at 12mA, since there is a large variation in Idss even for J-FETs of the same rank. By the way, in the #253 attached circuit diagram, there is only one BL in the GR rank. Is this a typo in the circuit diagram? Is there a deliberate purpose？
When measured without Å weighting, there was still a difference of about 8dB (depending on the low frequency cutoff of my phono amplifier and noise meter). It seems to be effective in improving this too.
The schematic is a bit of a work in progress, mostly completed but not perfectly so.
I tested the calibration circuit for models that represent a range of thresholds, roughly corresponding to the expected limits for the devices selected. The BL device has a higher IDSS and is there as a test and is but one of several going from all GR to all BL that I conducted to see if the calibration and noise performances seemed reasonably consistent.
The final version will be all GR, or at least that's the intent, but we'll see when it's built.
I tested the calibration circuit for models that represent a range of thresholds, roughly corresponding to the expected limits for the devices selected. The BL device has a higher IDSS and is there as a test and is but one of several going from all GR to all BL that I conducted to see if the calibration and noise performances seemed reasonably consistent.
The final version will be all GR, or at least that's the intent, but we'll see when it's built.
More commentary on the noise.
The actual dominant noise as far as the ultra LF is concerned may not be the equivalent "load" R on the input coupling cap, but the 1/f noise of the feedback opamp in the calibration loop. I alluded to the effect of the feedback opamp earlier.
This is proving very hard to simulate with "real" opamps, the TL072B model being one of the few that I can get to actually converge, and it has terrible 1/f noise.
I can get convergence with an ideal, noiseless opamp and the 20-20k unweighted output noise, with a 1uF/120k/1.2meg input combo drops from c.340uv to 145uv.
My opamp of choice for the calibration loop is the OPA1678, and with the OPA1656 replaced with an ideal opamp with similar AVol and GBWP the circuit does converge and gives a 145uv value.
Pragmatically, the answer is probably to go with an OPA1678 in the calibration loop and the OPA1656 in the transimpedance cell. and forget about having it simulatable.
The actual dominant noise as far as the ultra LF is concerned may not be the equivalent "load" R on the input coupling cap, but the 1/f noise of the feedback opamp in the calibration loop. I alluded to the effect of the feedback opamp earlier.
This is proving very hard to simulate with "real" opamps, the TL072B model being one of the few that I can get to actually converge, and it has terrible 1/f noise.
I can get convergence with an ideal, noiseless opamp and the 20-20k unweighted output noise, with a 1uF/120k/1.2meg input combo drops from c.340uv to 145uv.
My opamp of choice for the calibration loop is the OPA1678, and with the OPA1656 replaced with an ideal opamp with similar AVol and GBWP the circuit does converge and gives a 145uv value.
Pragmatically, the answer is probably to go with an OPA1678 in the calibration loop and the OPA1656 in the transimpedance cell. and forget about having it simulatable.
How about omitting OPA1678 from the simulation? Maybe you could replace it with the much less complicated "UniversalOpamp3" (found at the very end of the list of available opamp simulation models in LTSPICE) ... Maybe you'll get lucky, maybe the reduced circuit complexity will improve convergence in the simulator? Set the GBWP and SR and PM and RRIO and other parameters on the UniversalOpamp3 dropdown menu, then add fictitious external circuitry to produce the noise behavior you expect the real non-universal OPA1678 opamp to display?
Actually, I've been replacing all the opamps with universal opamp2 customized for each position and everything runs properly. However, I've never been too comfortable with that level of abstraction. I can, in fact, run the gain block with the "OPA1656" as the universal2 model and the OPA1678 in place and the hand calculations and the simulations are in good agreement.
I'm presently optimizing the interaction between all of the DC offset correction loops using universalopamp2 in all the locations.
I'm presently optimizing the interaction between all of the DC offset correction loops using universalopamp2 in all the locations.
I've attached what I hope is the "final" schematic. There may be some changes as the layout progresses, but this is what we are going with.
A single OPA1678 may be used for both channels, and if so, some minor changes will occur.
I'm finished with simulations, for now at least.
A single OPA1678 may be used for both channels, and if so, some minor changes will occur.
I'm finished with simulations, for now at least.
The layout is progressing, but slowly. A couple of volunteers have appeared to help accelerate the process, which is much appreciated.
More feedback from interested parties has been received, so a few changes have been made. A sub sonic filter- a 2nd order Butterworth high pass at c. 3.5Hz- has been added. This improves the low frequency PSRR and input referred "1/f noise" while reducing the settling time for the preamp so that a shorter, and much more acceptable, power on time delay can be had. It also removes the need for the DC feedback loop at the output.
The ability to have a volume control has been added so the unit can act as a preamp, if desired. It has no significant effect on the distortion or RIAA compliance.
The latest LTspice and Kicad7 schematics are attached. In practice I expect that the calibration loop OPA1678 opamps will be replaced by OPA2202s to further reduce the ultra LF noise, but I have yet to obtain the TI models although the universalopamp2 models suggest that it's a good choice.
More feedback from interested parties has been received, so a few changes have been made. A sub sonic filter- a 2nd order Butterworth high pass at c. 3.5Hz- has been added. This improves the low frequency PSRR and input referred "1/f noise" while reducing the settling time for the preamp so that a shorter, and much more acceptable, power on time delay can be had. It also removes the need for the DC feedback loop at the output.
The ability to have a volume control has been added so the unit can act as a preamp, if desired. It has no significant effect on the distortion or RIAA compliance.
The latest LTspice and Kicad7 schematics are attached. In practice I expect that the calibration loop OPA1678 opamps will be replaced by OPA2202s to further reduce the ultra LF noise, but I have yet to obtain the TI models although the universalopamp2 models suggest that it's a good choice.
The layout is still progressing, although slowly. It's a tricky one compared to the past efforts. The OPA1678s in the cal loop are now OPA202s which are single superbeta input opamps in a SOIC-8 package.
You've noticed the slew rate of the OPA202? I guess its OK for small signal levels, but it wouldn't work in a mixing desk very well(!)
The OPA202 is only used in the front-end calibration block in the DC feedback integrator and not for audio signal handling. The bandwidth of this stage is a fraction of a Hz. Hence the slew rate required is in the order of 1v/sec and the opamp only "slews" during the initial acquisition after power on.
The other opamps are OPA1678 (9v/us)- which is only used in a circuit with 150kHz bandwidth, and opa1656 (24v/us) which is used in the full bandwidth stages.
The other opamps are OPA1678 (9v/us)- which is only used in a circuit with 150kHz bandwidth, and opa1656 (24v/us) which is used in the full bandwidth stages.
Hmm, that's going to be OK then! - I guess the low power consumption is quite handy for such auxiliary circuitry.
Actually, the choice was made not as a power consideration (the total design takes c, 200mA from +/-15v supplies so a single opamp is largely irrelevant) but as a "1/f" consideration. The low 1/f corner and the low current noise of the OPA202 substantially improved the low frequency noise of the design, which otherwise can be pretty large due to the input coupling cap/input impedance of the JFET input gain stage (see prior posts).
The OPA202 also is low distortion at low gains/low frequency which helps to reduce the distortion injected into the input sufficiently to not degrade the overall distortion characteristic of the single ended input stage.
The goal was to make the calibration loop as acoustically transparent as possible- and a superbeta input low distortion opamp was just the ticket!
The OPA202 also is low distortion at low gains/low frequency which helps to reduce the distortion injected into the input sufficiently to not degrade the overall distortion characteristic of the single ended input stage.
The goal was to make the calibration loop as acoustically transparent as possible- and a superbeta input low distortion opamp was just the ticket!
Layout progressing.
Inputs at the top left, outputs at the middle left, AC input from the wall wart at the bottom left. Board c. 290mm x 250mm. Still a few changes to be made.
Inputs at the top left, outputs at the middle left, AC input from the wall wart at the bottom left. Board c. 290mm x 250mm. Still a few changes to be made.
The "test" boards (see above), which are the, hopefully, fully functional design, have been ordered, as will be the Mouser cart (today). The boards are c. $10 each plus shipping in quantities of 5.
Provisions have been added for the projected IR remote control/front panel board, but that is not essential for operation.
The Mouser cart (all except case/sockets/wall wart) is $124 including shipping.
My collaborator, Bill Hirsch, who did the layout, will be assembling the board and I'll test/debug it and post the results.
I (finally) have some measurement results. The unit is not in a box, but some attempt has been made to shield it to get the mains hum components down. Those darn input FETs.
The results include the "loop back" distortion of the RME ADC/DAC.
c. 50mdB p-p RIAA accuracy 20Hz-20kHz.
Noise floor with 10 ohm source impedance.
The 10kHz distortion has degraded a tad due to the increase in "compensation" capacitance for the transimpedance cell.
Everything works as expected.
At some point I'll measure the results with the other gain settings, but I'm done for now.
The results include the "loop back" distortion of the RME ADC/DAC.
c. 50mdB p-p RIAA accuracy 20Hz-20kHz.
Noise floor with 10 ohm source impedance.
The 10kHz distortion has degraded a tad due to the increase in "compensation" capacitance for the transimpedance cell.
Everything works as expected.
At some point I'll measure the results with the other gain settings, but I'm done for now.
A lower cost, lower complexity version has been requested. Now that the more complicated design has been validated, here it is.
Single stereo input, MC/MM, 35/41/44/50/55/61/645/70dB gain. MC/MM loads switchable. Simple single pole "crossfeed' (warp), <100mdB RIAA error 20Hz-20kHz, 3.5Hz 2nd order high pass filter, Mono/stereo switching. Same noise/distortion performance as the more complicated one.
Output turn on/off relay with extra protection against loss of power and multiple on/off cycles.
Single stereo input, MC/MM, 35/41/44/50/55/61/645/70dB gain. MC/MM loads switchable. Simple single pole "crossfeed' (warp), <100mdB RIAA error 20Hz-20kHz, 3.5Hz 2nd order high pass filter, Mono/stereo switching. Same noise/distortion performance as the more complicated one.
Output turn on/off relay with extra protection against loss of power and multiple on/off cycles.
Mono/stereo switching is needed.
Is more switching part's complicated to handle distortion / SNR.
It's just a thought
Is more switching part's complicated to handle distortion / SNR.
It's just a thought
The noise/distortion is unchanged. There is just a degradation in warp filter performance (going from 2nd order to 1st order degrades the stereo frequency response "flatness" and degrades the 1kHz channel separation), the removal of balanced output, the removal of the dual input capability, and going to mechanical panel mounted switches rather than relays.