First stage:
1596R+47n >> best for 2121,73Hz
Second stage:
3k3 with 330k >> gain for DC 40dB. So for 1kHz will be 20dB less.
330k in NFB with 9,5nF gives 50,77Hz. 10n as the closest value >> 48,23Hz.
(330k/9)-3300=33366R >> 33k as the closest value. With 10n gives 482,29Hz.
3k3 with 10u cuts the lowest frequency.
1596R+47n >> best for 2121,73Hz
Second stage:
3k3 with 330k >> gain for DC 40dB. So for 1kHz will be 20dB less.
330k in NFB with 9,5nF gives 50,77Hz. 10n as the closest value >> 48,23Hz.
(330k/9)-3300=33366R >> 33k as the closest value. With 10n gives 482,29Hz.
3k3 with 10u cuts the lowest frequency.
So, I had made the mistake of entering 300.... however, the best numbers from the spreadsheet I get are:
R1=1K596
C1=0.047u
R2=R3=3K3
R4=250K
R5=33K
C2=0.0056u
R6=250K
C3=47uf OR MORE.
This results in a sum of squared errors of the Real part of the voltage as below 10, for 355 sample points. The biggest error is at 20 Hz, 0.72 db down from standard. This is back on track at 44Hz.
Now of course, my spreadsheet could be wrong. But the caculation for each element was verified.
I will be trying to make a measurement of the standard tonight. Then will make the adjustment and remeasure.
R1=1K596
C1=0.047u
R2=R3=3K3
R4=250K
R5=33K
C2=0.0056u
R6=250K
C3=47uf OR MORE.
This results in a sum of squared errors of the Real part of the voltage as below 10, for 355 sample points. The biggest error is at 20 Hz, 0.72 db down from standard. This is back on track at 44Hz.
Now of course, my spreadsheet could be wrong. But the caculation for each element was verified.
I will be trying to make a measurement of the standard tonight. Then will make the adjustment and remeasure.
I check it directly against riaa curve. Sorry, I wrote 355 sample points. but I have only 83, which is plenty.
I found a couple other mistakes in my spreadsheet, but nothing major. If you want to make such calculations, make sure to use the atan2 function in the spreadsheets, not atan.
My education includes a lot of statistics, so minimizing the sum of squares error is what I am doing -- I am not really worried about anything else.
When I use the standard values, I calculate the maximum deviation from RIAA to be about 3.6 db. When I use my values, it is 0.7 db.
I have never seen any claims from 47 labs about how close their product matches the RIAA curve.
I have just read on Arthur Salvatore's site, that the first stage OP amp is an OP27. I can agree with this choice. It is better than opa627BP -- less noise.
I found a couple other mistakes in my spreadsheet, but nothing major. If you want to make such calculations, make sure to use the atan2 function in the spreadsheets, not atan.
My education includes a lot of statistics, so minimizing the sum of squares error is what I am doing -- I am not really worried about anything else.
When I use the standard values, I calculate the maximum deviation from RIAA to be about 3.6 db. When I use my values, it is 0.7 db.
I have never seen any claims from 47 labs about how close their product matches the RIAA curve.
I have just read on Arthur Salvatore's site, that the first stage OP amp is an OP27. I can agree with this choice. It is better than opa627BP -- less noise.
I tried 1K596 by paralleling a 100K Vishay S102 0.002% resistor with the 1K62 Vishay nude. It measured perfect 1K596!
I do not like this sound -- much too hard sounding. After listening, I consulted the spreadsheet, which indicated that ALL of the frequencies above 1K will get a slight boost -- only 0.1 db -- which I thought should be irrelevant. The sum of the squares of the error increase to about 265 from about 255. My spreadsheet suggests that with different values in the second stage, this sum of squares of the error can get under 10!
Still waiting for parts to arrive.
I do not like this sound -- much too hard sounding. After listening, I consulted the spreadsheet, which indicated that ALL of the frequencies above 1K will get a slight boost -- only 0.1 db -- which I thought should be irrelevant. The sum of the squares of the error increase to about 265 from about 255. My spreadsheet suggests that with different values in the second stage, this sum of squares of the error can get under 10!
Still waiting for parts to arrive.
Last edited:
The boutique parts shipped today, but I could not help myself. I did the modification with some cheap 10 cent styrene caps - but really good resistors.
The minimum changes are:
R4 = 250K
C2 = 0.0056uF = 5.6 nf = 5600 pf
And it sounds terrific!
According to my spreadsheet, this now has a sum of squares error from RIAA of 9.25 measured at 83 different frequencies. The sum of squares error from RIAA started at 255 over the same 83 frequencies.
It will only get better once I find a teflon 0.0056 uF cap....I am a happy guy!
-----
It will take a bit of effort to voice this properly. R4 is likely a little too small. 300K is probably about right, unless you are willing to change to C3 to 47uF (bipolar). Voicing will happen after New Years. It is clear to me that there are errors in RIAA conformance... it's just a matter of how much and where you can stand them.
The minimum changes are:
R4 = 250K
C2 = 0.0056uF = 5.6 nf = 5600 pf
And it sounds terrific!
According to my spreadsheet, this now has a sum of squares error from RIAA of 9.25 measured at 83 different frequencies. The sum of squares error from RIAA started at 255 over the same 83 frequencies.
It will only get better once I find a teflon 0.0056 uF cap....I am a happy guy!
-----
It will take a bit of effort to voice this properly. R4 is likely a little too small. 300K is probably about right, unless you are willing to change to C3 to 47uF (bipolar). Voicing will happen after New Years. It is clear to me that there are errors in RIAA conformance... it's just a matter of how much and where you can stand them.
Last edited:
The parts arrived, so I put them in immediately.
R4 = 275K ( 249K caddock + 1% metal film (mistake ordering)) was 330K
C2 = 0.0056 uF Relcap RTE (was 0.01)
C3 = 47uF Elna RBD
Results are terrific even without break in! Of course, since R4 is now below 300K a Vishay nude can also be installed. A Russian teflon is on the way for the Relcap.
R4 = 275K ( 249K caddock + 1% metal film (mistake ordering)) was 330K
C2 = 0.0056 uF Relcap RTE (was 0.01)
C3 = 47uF Elna RBD
Results are terrific even without break in! Of course, since R4 is now below 300K a Vishay nude can also be installed. A Russian teflon is on the way for the Relcap.
But the point of this is that given the design of a current to voltage convertor followed by a difference amp is not hard to model. The equations are closed form solutions. So is Riaa. Then, just take the results of the model, normalize to 1k, subtract normalized Riaa values, and square these errors.
This yields a sum of squared errors of 258 for the original design. The above changes lower the sum of squares of the errors to 12.5. I would think the changed implementation would be closer to Riaa regardless of whether or not specific time constants are met.
Using. 250K r4 gives an even closer sum of squared error = 8.34. How can this be a bad thing?
This yields a sum of squared errors of 258 for the original design. The above changes lower the sum of squares of the errors to 12.5. I would think the changed implementation would be closer to Riaa regardless of whether or not specific time constants are met.
Using. 250K r4 gives an even closer sum of squared error = 8.34. How can this be a bad thing?
I found the error in my spreadsheet.
With my values, the sum of squares error is 145.
With the stock values, it is 1.9. With R1=1K596 it is 1.87.
The circuit values are correct and cannot be improved in any significant way.
R4 to 300K gives sum of squared error of 0.58! 310k gives 0.41!!!
With my values, the sum of squares error is 145.
With the stock values, it is 1.9. With R1=1K596 it is 1.87.
The circuit values are correct and cannot be improved in any significant way.
R4 to 300K gives sum of squared error of 0.58! 310k gives 0.41!!!
Last edited:
After playing around, the best values I could generate from the spreadsheet were:
C3=10uF
R4=300000
R5 = 32600 ( 32570 == 33000 || 2M5 )
C2 = 0.0099uF
This is very very close to RIAA. Important frequencies are within 0.05db,with frequency extremes being 0.08db out.
This has not been tested.... parts are coming to try to build up the 0.0099 uF cap. The change of R4 and R5 will get closer... the 2M5 resistors are also on the way.,
C3=10uF
R4=300000
R5 = 32600 ( 32570 == 33000 || 2M5 )
C2 = 0.0099uF
This is very very close to RIAA. Important frequencies are within 0.05db,with frequency extremes being 0.08db out.
This has not been tested.... parts are coming to try to build up the 0.0099 uF cap. The change of R4 and R5 will get closer... the 2M5 resistors are also on the way.,