OK, so it all seems to indicate that there is capacitive crosstalk from the left channel to the input of the right channel.
Possible receiving parts: R13, C2, R6, R3, C4, right-channel positive op-amp input, all wiring and copper islands and the connector that are connected to either side of C2.
Possible transmitting parts: anything connected to a left-channel node with a signal on it, probably an amplified signal:
Left-channel op-amp output, R40, R70, R80, R90, C60, C50 (side that is not connected to the negative op-amp input), R110, and their connecting wires, connector and copper islands.
On the photograph, the right channel input cable gets close to R70, but only to the side of R70 that is connected to the negative input of the left-channel op-amp, which carries a non-amplified signal. The dampers are soldered straight to the input connectors and the right-channel input connector is adjacent to the left-channel output connector. Does cutting out R3 and C4 or putting a shield between the left output and right input connector help?
If you should have a couple of capacitive crosstalk paths that are about equally strong, eliminating one with a shield will only give a minor improvement. For example, suppose there is capacitive coupling from the wires of R70, R80, R90 to C2 and about the same amount from the left output connector and its resistor to the right input connector and its damper. You then have to shield both to get a substantial improvement.
Possible receiving parts: R13, C2, R6, R3, C4, right-channel positive op-amp input, all wiring and copper islands and the connector that are connected to either side of C2.
Possible transmitting parts: anything connected to a left-channel node with a signal on it, probably an amplified signal:
Left-channel op-amp output, R40, R70, R80, R90, C60, C50 (side that is not connected to the negative op-amp input), R110, and their connecting wires, connector and copper islands.
On the photograph, the right channel input cable gets close to R70, but only to the side of R70 that is connected to the negative input of the left-channel op-amp, which carries a non-amplified signal. The dampers are soldered straight to the input connectors and the right-channel input connector is adjacent to the left-channel output connector. Does cutting out R3 and C4 or putting a shield between the left output and right input connector help?
If you should have a couple of capacitive crosstalk paths that are about equally strong, eliminating one with a shield will only give a minor improvement. For example, suppose there is capacitive coupling from the wires of R70, R80, R90 to C2 and about the same amount from the left output connector and its resistor to the right input connector and its damper. You then have to shield both to get a substantial improvement.
Apart from your other recommendations, what about pins 1 and 5 of the 627? Trim pins..? I can cut them flush with the body. There's no "If not used..." guidance in the datasheet.
I think they should be fairly harmless. The signal voltage swing on the left channel's offset trimming pins should be much smaller than on the left channel output, while the offset trimming pins and their copper islands are further from the right channel. The right channel offset trimming pins are also far less sensitive to injected interfering currents than the open input.
Still working on the perf-board prototype. PCB version is further out, unfortunately.
All good fortune,
Chris
All good fortune,
Chris
New things done to-date:
1) Bridged all unused 5-position breadboard strips to the center backside Gnd.
2) Isolated (dismounted) the Right input RCA connector; wrapped it in a Ziplock bag and closed the lid. No plug.
3) Separated the Left and Right channel circuitry with grounded copper tape in an attempt to form a barrier between C80A&B and R13/C2/R6.
4) Removed U2.
5) Removed the R&L input dampers R3/C4 and R30/C40.
Impressions:
The combo of #1, 3 and 5 may have provided the attached improvements. #2 gained nothing. #4 did nothing good. I did #1 and #5 at the same time.😡 #3 maybe helped.
Attached are:
1) One pic of the previous “No plug Ghost 6-25-25”
2) Three current pics: “No plug Ghost”, “1K plug Ghost” and “Close-up 1K plug Ghost”.
3) Two new and different pics of the square wave as input.
4) Four pics now with the Right channel driven: “No plug - Left”, “1K plug - Left”, “Close-up no plug - Left” and Close-up 1K plug – Left.
5) A number of pics of the attempt to separate U2 Left from U1 Right componentry with 1” wide conductive copper adhesive tape. The tape exactly bisects the circuitry from board level up to a 1” high shield.
Observations re attached pics 1 - 4.
1 & 2: The current vs. last week’s “No plug Ghost” are dramatically improved, resp. A trace of 1kHz sine remains in the current Ghost.
2: “1K plug Ghost” and “Close-up 1K plug Ghost” are the cleanest yet without a trace of the 1KHz Left signal.
3: The square wave Ghost has lost much definition and sharpness. Less obvious.
4: All of these “golden” Right-driven, Left Ghost shots are nearly equaled by those from 2), above.
5) The U1/U2 circuits shield was the best I could come up with. ("B-F" means - From back to front, in-order. )
Question:
1) Looking at the "1K plug" Right and Left pics what level of hash noise is explainable as in self-generated noise from Rs and the op amp or the scope?
1) Bridged all unused 5-position breadboard strips to the center backside Gnd.
2) Isolated (dismounted) the Right input RCA connector; wrapped it in a Ziplock bag and closed the lid. No plug.
3) Separated the Left and Right channel circuitry with grounded copper tape in an attempt to form a barrier between C80A&B and R13/C2/R6.
4) Removed U2.
5) Removed the R&L input dampers R3/C4 and R30/C40.
Impressions:
The combo of #1, 3 and 5 may have provided the attached improvements. #2 gained nothing. #4 did nothing good. I did #1 and #5 at the same time.😡 #3 maybe helped.
Attached are:
1) One pic of the previous “No plug Ghost 6-25-25”
2) Three current pics: “No plug Ghost”, “1K plug Ghost” and “Close-up 1K plug Ghost”.
3) Two new and different pics of the square wave as input.
4) Four pics now with the Right channel driven: “No plug - Left”, “1K plug - Left”, “Close-up no plug - Left” and Close-up 1K plug – Left.
5) A number of pics of the attempt to separate U2 Left from U1 Right componentry with 1” wide conductive copper adhesive tape. The tape exactly bisects the circuitry from board level up to a 1” high shield.
Observations re attached pics 1 - 4.
1 & 2: The current vs. last week’s “No plug Ghost” are dramatically improved, resp. A trace of 1kHz sine remains in the current Ghost.
2: “1K plug Ghost” and “Close-up 1K plug Ghost” are the cleanest yet without a trace of the 1KHz Left signal.
3: The square wave Ghost has lost much definition and sharpness. Less obvious.
4: All of these “golden” Right-driven, Left Ghost shots are nearly equaled by those from 2), above.
5) The U1/U2 circuits shield was the best I could come up with. ("B-F" means - From back to front, in-order. )
Question:
1) Looking at the "1K plug" Right and Left pics what level of hash noise is explainable as in self-generated noise from Rs and the op amp or the scope?
Attachments
-
No Plug Ghost 6-25-25.JPG -
Shield End 2.JPG -
Shield End 1.JPG -
B-F; R13,C2,Shld,R6,Jmp-pin3.JPG -
Shield Top.JPG -
Shield overview.JPG -
Close-up 1K plug - Left.JPG -
Close-up no plug - Left.JPG -
1K plug - Left.JPG -
No plug Ghost.JPG -
1K plug Ghost.JPG -
Close-up 1K plug Ghost.JPG -
No plug Square Ghost 1.JPG -
No plug Square Ghost 2.JPG -
No plug - Left.JPG
Rough estimate:
A RIAA correction curve is similar to a first-order low-pass at 50 Hz. The noise bandwidth of a first-order low-pass is π/2 times the cut-off frequency, so that's about 80 Hz. You lose about 16 Hz due to the subsonic filter, but the fact that the transfer is flat between 500 Hz and 2122 Hz at about 1/10 of the low-frequency gain compensates for that. All in all, 80 Hz with respect to the gain of about 2000 times that you have between 16 Hz and 50 Hz.
The typical input noise voltage density of the OPA627 is somewhere between 4.8 nV/√Hz and 7.5 nV/√Hz at these frequencies, say about 6 nV/√Hz. The 1 kohm resistor produces about 4 nV/√Hz. All other contributions are much smaller (for this specific case with 1 kohm source and no A- or ITU-R 468-weighting).
The root of the sum of the squares is roughly 7.2 nV/√Hz. The root of the noise bandwidth is almost 9 √Hz, so you have about 65 nV RMS times the low-frequency gain of 2000, or 130 uV RMS at the output. Assuming a Gaussian distribution, the quasi peak-peak value is about 6 times more, 780 uV quasi peak-peak.
A RIAA correction curve is similar to a first-order low-pass at 50 Hz. The noise bandwidth of a first-order low-pass is π/2 times the cut-off frequency, so that's about 80 Hz. You lose about 16 Hz due to the subsonic filter, but the fact that the transfer is flat between 500 Hz and 2122 Hz at about 1/10 of the low-frequency gain compensates for that. All in all, 80 Hz with respect to the gain of about 2000 times that you have between 16 Hz and 50 Hz.
The typical input noise voltage density of the OPA627 is somewhere between 4.8 nV/√Hz and 7.5 nV/√Hz at these frequencies, say about 6 nV/√Hz. The 1 kohm resistor produces about 4 nV/√Hz. All other contributions are much smaller (for this specific case with 1 kohm source and no A- or ITU-R 468-weighting).
The root of the sum of the squares is roughly 7.2 nV/√Hz. The root of the noise bandwidth is almost 9 √Hz, so you have about 65 nV RMS times the low-frequency gain of 2000, or 130 uV RMS at the output. Assuming a Gaussian distribution, the quasi peak-peak value is about 6 times more, 780 uV quasi peak-peak.
Whoa - slow down...
So Marcel, by "quasi peak-peak" a level is determined by injecting a signal into a circuit like the attached "Q-P circuit"? Or is there a distinction between Q-P and Q-P-P? Google doesn't seem to recognize quasi peak-peak. All of my measurements are in Vpp. When you say a rough estimate of 780uV Qpp that means for what - broadband noise, pink or white..? Is that 780uV Qpp noise viewable on a scope? Is that what I'm seeing in the "1K plug" screen shots? Should it be divided by 2 to get the "peak" value????
Attachments
I meant an estimate of the value you should see on an oscilloscope as the difference between the positive and negative peaks. Spectrally, the noise is strongest between 16 Hz and 50 Hz because the gain of the RIAA amplifier is highest in that frequency range.
Noise usually (though not necessarily) has a Gaussian distribution, the RMS value is the standard deviation. Gaussian random signals stay between -3 and +3 standard deviations most of the time, so 6 standard deviations is a reasonable rule of thumb for the value you will read off from an oscilloscope. It's not a real peak-to-peak value though, because the tails of a Gaussian distribution extend indefinitely.
Given the RIAA amplifier bandwidth, the probability distribution and some very rough guesstimates, I expect that the noise you measure will exceed +3 or -3 standard deviations about once every 22 seconds, and +8 or -8 standard deviations about once every 1.5 million years.
Noise usually (though not necessarily) has a Gaussian distribution, the RMS value is the standard deviation. Gaussian random signals stay between -3 and +3 standard deviations most of the time, so 6 standard deviations is a reasonable rule of thumb for the value you will read off from an oscilloscope. It's not a real peak-to-peak value though, because the tails of a Gaussian distribution extend indefinitely.
Given the RIAA amplifier bandwidth, the probability distribution and some very rough guesstimates, I expect that the noise you measure will exceed +3 or -3 standard deviations about once every 22 seconds, and +8 or -8 standard deviations about once every 1.5 million years.
You are playing with my non-mathematical head now Marcel. Might need a narrow band pass filter. But stay tuned b/c all of this may be moot. I'll show you all something that will maybe have even Marcel scratching his head.
Sorry for the delay.
At the end of my last post #650 I thought it time to put the box into the main system and scope it there.
The TT is the full-auto Mits LT-30 with a nice clean muting switch. As for photo titles the basic translations are:
“5KHz Test – Right” shows the Right channel’s 5KHz LP test signal output in Yellow where the Blue trace is the cartridge’s Left crosstalk output. Here the R→L channel separation is a decent 27.9dB. This is aka the Arm Down, play position.
“Arm UP – shorted” means that after the 5KHz test, the tonearm was lifted; is stationary in the non-playing position where the TT has each channel’s two ± coil wires independently shorted together. In order to see how clean the pre’s output was with presumably 0V inputs the scope’s horizontal time base was changed to 20mS/div and the vertical voltage resolution increased to 2mV/div and … What the hell..? The “spike” amplitude measured 12.3mVp-p.
“Arm UP – shorted_2” = the time base is now 4mS/div where two spikes are caught arriving 29.7mS apart (33.6Hz).
“Arm UP – High freq zoom” = here I zeroed-in on the patch of hi-freq hash at the 2nd spike’s trailing edge, above. The time-base is now only 20nS/div and a “cycle” measured 37.7MHz.
“Arm UP – High freq zoom_2” = same patch of hash but expanded to 8nS/div. Here a cycle measured at 43.8MHz.
“Arm UP – High freq zoom_3” = a different capture at 8nS/div measured 42.0MHz.
“Arm Down – Rest Open” means the TT has opened the cartridge shorts b/c the arm is in the Play position or here, sitting down on the end-of-play arm rest. IOW, Arm Down is what gets mixed onto the music signals.
Now what? 😢
At the end of my last post #650 I thought it time to put the box into the main system and scope it there.
The TT is the full-auto Mits LT-30 with a nice clean muting switch. As for photo titles the basic translations are:
“5KHz Test – Right” shows the Right channel’s 5KHz LP test signal output in Yellow where the Blue trace is the cartridge’s Left crosstalk output. Here the R→L channel separation is a decent 27.9dB. This is aka the Arm Down, play position.
“Arm UP – shorted” means that after the 5KHz test, the tonearm was lifted; is stationary in the non-playing position where the TT has each channel’s two ± coil wires independently shorted together. In order to see how clean the pre’s output was with presumably 0V inputs the scope’s horizontal time base was changed to 20mS/div and the vertical voltage resolution increased to 2mV/div and … What the hell..? The “spike” amplitude measured 12.3mVp-p.
“Arm UP – shorted_2” = the time base is now 4mS/div where two spikes are caught arriving 29.7mS apart (33.6Hz).
“Arm UP – High freq zoom” = here I zeroed-in on the patch of hi-freq hash at the 2nd spike’s trailing edge, above. The time-base is now only 20nS/div and a “cycle” measured 37.7MHz.
“Arm UP – High freq zoom_2” = same patch of hash but expanded to 8nS/div. Here a cycle measured at 43.8MHz.
“Arm UP – High freq zoom_3” = a different capture at 8nS/div measured 42.0MHz.
“Arm Down – Rest Open” means the TT has opened the cartridge shorts b/c the arm is in the Play position or here, sitting down on the end-of-play arm rest. IOW, Arm Down is what gets mixed onto the music signals.
- The Arm UP signal was a total surprise and is audible like a low-frequency tapping or popping.
- The Arm Down signal is not great. There’s a trace of 60Hz hum in the Right and the overall amplitudes are at best, ~1mV with the cart’s 800 DCR coils across the inputs. I suspect the hash to have the same-ish content as the Arm UP examples.
| Coil Impedance | 2,700 ohms (1 kHz) |
| DC Resistance | 800 ohms |
| Coil Inductance | 460 mH (1 kHz) |
| Output Voltage | 4.0 mV (at 1 kHz, 5 cm/sec) |
Now what? 😢
Attachments
Is this with or without the RC dampers? 37 MHz...44 MHz is close to the estimated quarter-wave transmission line resonance of the cable of the turntable. In any case it's strange, as you never needed the dampers for the OPA627.
This is without the RC dampers. The TT cable is also shorter this time and is not the Blue Jeans LC_1. It's 3ft now and was either 1m or 4ft with the previous Mits LT-20 TT. Also the total C right up through the arm is now ~ 165pF. In those days of whop-thump I don't think I put idle states under such scrutiny as now. IOW I never zoomed-in like with these current pics. Usually when looking at crosstalk I would have the smooth wall channel set to say, 20mV, maybe 10mV or 50mV and the driven wall 1V. Where do you think the recurring low-frequency negative-going "daggers" are coming from? The Arm UP waveforms are continuous vs. being triggered by the original scratchy slide switch, btw. These dance on the screen and make a puttering sound.
- I'll put the dampers back in.
- Short the input caps again.
- Change carts
These pics show the behavior if the Marvel Pre with the 220ohm/15pF dampers re-installed right across R6 at the + inputs of U1 and U2. The cart specs are the same as in post #652, above.
The two "Whoppers" 1 and 2 showed up early in the Rt chan but did not seem to repeat. Note how the Rt Whopper dwarfs the Lt counterpart in amplitude. One other pic shows a "normal" shot of the stronger right vs. the left which seemed to be the norm. So I reversed R&L TT cables and now the Lt became dominant as far as amplitude and activity. Lastly there's a big one on the Lt chan also with the TT cables reversed (i.e. the Right chan did that).
- The Damper made no final difference. This time it was installed at the doorsteps of U1 and U2.
- There is something different between the Right and Left cart inputs.
- The OPA627 is unhappy in some way.
Attachments
Chris: We all need to see a 2nd build. My problems can't be all that unique. Your results would be most appreciated.
I might suggest both that we are building different versions (I'm making a a dead-bug perf-board prototype with DIL8 OPA2134 and other parts on hand that I can still see good enough to work) and that a general-enough version might be worth the community effort to develope. Obviously, the 6u8F capacitors and their signal loops will dominate the PCB design.
I'd hoped in my prototype to pretend to ignore power supply issues by powering by 4x 9Volt batteries, a DPST switch and a pair of series'd red LEDs. Based on your experience, and an inate aversion to being on the cutting edge, a 50MHz GBP version is just beyond my grasp. It would have to be already soldered into the PCB - maybe I could do it, maybe not - but not trivial.
For me the only important thing is the topology. I have three Advent 300 receivers (my personal favorite audio electronics) and am rapidly losing my hearing. Beauty remains important. Maybe most important.
Americans stand up for democracy,
Chris
I'd hoped in my prototype to pretend to ignore power supply issues by powering by 4x 9Volt batteries, a DPST switch and a pair of series'd red LEDs. Based on your experience, and an inate aversion to being on the cutting edge, a 50MHz GBP version is just beyond my grasp. It would have to be already soldered into the PCB - maybe I could do it, maybe not - but not trivial.
For me the only important thing is the topology. I have three Advent 300 receivers (my personal favorite audio electronics) and am rapidly losing my hearing. Beauty remains important. Maybe most important.
Americans stand up for democracy,
Chris
I have built 2 versions of Marcel's preamp. One 2nd order HPF version with an NE5532 on an old Philips PCB and another 3rd order HPF version with an NJM2068 using the topology and repurposing the PCB in my integrated amp. Then there is Ansible's version who's applied an OPA1656. He's even made a PCB design (switchable 2nd / 3rd order / input damper) and his version works well too. Perhaps he wants to share his Gerbers? It seems that either the OPA627 or your layout is causing you trouble. Feels like self-flagellation to me continuing this route.