Sigurd Ruschkow said:Good work, Syn08!
I guess you are now in the 80dB 20-20k, flat, MC-RIAA club
What kind of eq total input noise 20-20k do you get incl RIAA transfer function?
BTW,
your matched JFETs, what noise did you select for for the SJ74 and SK170?
I am glad to read that you will make it signal cap less! and also as dual mono! For me, that is the only way.
Problem with BF862 is that there is no P-ch compl part.
83dB SNR unweighted, 0.5mV input signal, 64dB gain.
Preliminary, >92dB SNR RIAA, 0.5mV input signal, 64dB gain.
Sorted 0.6-0.7nV/rtHz JFETs @1KHz, Id=10mA, Vds=5V. Not easy to find (in particular for 2SK170), but the recent purchase of 200pcs. of 2SK170BL from the group buy around helped.
Dual mono is a luxury, not even sure if I'll keep it in the final release.
It's not will it is cap less
I'll post the whole schematic soon.2SJ74BL is not really required for ultra low noise design. While it's certainly nice to use complementary pairs, to me the key to under 0.2nV/rtHz is either a) finding a way to parallel more JFETs while keeping the thing stable, or b) cooling down. I'm looking into both... (look at this http://www.cui.com/pdffiles/CP85238.pdf ).
syn08 said:
I wasn't suggesting the actual circuit, I posted it to show the topology, which could be made with jfets just the same. I don't think the noise contribution of the current source is an insurmountable problem either.
Without complementary input devices, +3dB more noise over single ended is the trade off you pay for a much better PSRR (as, like, opposed to that previous AudioExpress thingie with 0dB PSRR).
As for the 0.5nV noise spec, that doesn't include the feedback network. The schematic is from the MAT03 datasheet and follows a 0.32nV version.
Cheers,
Glen
Edit:
What SOT-23 complement have you found for the BF862?
I have only reached 81 dB SNR re 500uVrms 20-20k flat. 83 dB is excellent!
Luxuries are there enjoy
I also looked into cooling down the amp, but gave that up as it was too cumbersome. The overclocking guys use liquid nitrogen to cool down the CPU when setting records. Maybe that is one way to do it. Albeit cumbersome.
Sigurd
Luxuries are there enjoy
I also looked into cooling down the amp, but gave that up as it was too cumbersome. The overclocking guys use liquid nitrogen to cool down the CPU when setting records. Maybe that is one way to do it. Albeit cumbersome.
Sigurd
syn08 said:
83dB SNR unweighted, 0.5mV input signal, 64dB gain.
Preliminary, >92dB SNR RIAA, 0.5mV input signal, 64dB gain.
Sorted 0.6-0.7nV/rtHz JFETs @1KHz, Id=10mA, Vds=5V. Not easy to find (in particular for 2SK170), but the recent purchase of 200pcs. of 2SK170BL from the group buy around helped.
Dual mono is a luxury, not even sure if I'll keep it in the final release.
It's not will it is cap less
2SJ74BL is not really required for ultra low noise design. While it's certainly nice to use complementary pairs, to me the key to under 0.2nV/rtHz is either a) finding a way to parallel more JFETs while keeping the thing stable, or b) cooling down. I'm looking into both... (look at this http://www.cui.com/pdffiles/CP85238.pdf ).
syn08 said:
83dB SNR unweighted, 0.5mV input signal, 64dB gain.
Preliminary, >92dB SNR RIAA, 0.5mV input signal, 64dB gain.
Sorted 0.6-0.7nV/rtHz JFETs @1KHz, Id=10mA, Vds=5V. Not easy to find (in particular for 2SK170), but the recent purchase of 200pcs. of 2SK170BL from the group buy around helped.
Dual mono is a luxury, not even sure if I'll keep it in the final release.
It's not will it is cap less
2SJ74BL is not really required for ultra low noise design. While it's certainly nice to use complementary pairs, to me the key to under 0.2nV/rtHz is either a) finding a way to parallel more JFETs while keeping the thing stable, or b) cooling down. I'm looking into both... (look at this http://www.cui.com/pdffiles/CP85238.pdf ).
Excellent stuff. FYI amateur astronomers are already using triple stacked peltiers to cool down their CCD's. You might find some ideas there, condensation becomes a real problem. My calculations show 32 BF862's could do .135nV at 300K
The BF 862 is specified at 1 MHz. Its not clear that the performance will be there at audio frequencies.
The work involved in making a transformerless low noise input is impressive. And these are very good numbers.
At what point are you no longer getting a benefit from lower noise? The DCR of the cartridge coil is the first limit. They aren't that low anymore. The early Ortofon's were in the 3 Ohm range but 30-100 Ohms seems common now. And the output seems higher proportional to coil resistance. Better magnet technology or magnetic circuits? However the disk process itself has a noise floor before even the cutting amp is turned on, and its effectively pretty high. And analog tape has a well understood noise floor. Even the microphone noise floor could be a major limitation.
Perhaps a "quant" here would do the numbers to see how low an input referred noise level is required before there is little to be gained.
The work involved in making a transformerless low noise input is impressive. And these are very good numbers.
At what point are you no longer getting a benefit from lower noise? The DCR of the cartridge coil is the first limit. They aren't that low anymore. The early Ortofon's were in the 3 Ohm range but 30-100 Ohms seems common now. And the output seems higher proportional to coil resistance. Better magnet technology or magnetic circuits? However the disk process itself has a noise floor before even the cutting amp is turned on, and its effectively pretty high. And analog tape has a well understood noise floor. Even the microphone noise floor could be a major limitation.
Perhaps a "quant" here would do the numbers to see how low an input referred noise level is required before there is little to be gained.
Sigurd Ruschkow said:
Use this model (romove the Beta parametrization if you can't accomodate it). According to Philips, BF862 has a 0.5ohm gate resistor. This model also accounts for the SOT23 case parasitics, good for stability evaluation.
*DEVICE=BF862, NJF
.SUBCKT BF862 1 2 3
+ params:
+ BETA=47.80m
JBF862 4 7 6 J_BF862
Ld 1 4 1.1n
Ls 3 6 1.25n
Lg 2 5 0.78n
Rg 5 7 0.535
Cds 1 3 0.0001p
Cgs 2 3 1.05p
Cgd 1 2 0.201p
Co 4 6 0.35092p
*
.model J_BF862 NJF(Beta={BETA} Betatce=-.5 Rd=.8 Rs=7.5000
+ Lambda=37.300E-3 Vto=-.57093 Vtotc=-2.0000E-3 Is=424.60E-12
+ Isr=2.995p N=1 Nr=2 Xti=3 Alpha=-1.0000E-3 Vk=59.97
+ Cgd=7.4002E-12 M=.6015 Pb=.5 Fc=.5 Cgs=8.2890E-12 Kf=87.5E-18
+ Af=1)
.ENDS BF862
Sigurd Ruschkow said:Lyra's cartridges have about 5 Ohms DCR.
Still, the LP's noise is around 60-70 dB, so actually no need to make a 80 dB MC RIAA amp.
However, it is also the challenge to make super low noise amps that are part of the reason to make one
If you examine the spectra real time, the actual noise floor is lower than you think. The RMS sum of the impairments give an overly pessimistic number. I have many LP's where the the noise drops noticably between cuts (tape noise is worse).
I use Grado exclusively (400 Ohms) so this is just a fun exercise for me.
john curl said:
Hi John,
Can you post an impedance curve for the Watt1's?
Did you measure that 0.5 ohm at 2 kHz, or was it specified somewhere or maybe in a review?
Did Stereophile ever do a review on the Watt1's.
I'm fascinated by a speaker that has this wild an impedance curve.
Thanks,
Bob
What SYN-08 has shown here is what can be done to make an ultra low noise MC FET input phono stage.
This number has been achieved before by Japanese designs back in the late '70's and early 80's using bipolar devices, but they never sounded as good as a fet design. It might be RFI sensitivity and dynamic range that makes the difference.
This topic still remains inside the Blowtorch thread, because an earlier version of this input stage is in the Blowtorch picture (11) and also is in my own personal unit.
I made two mistakes in the original design that is more than 25 years old now. One important one was where I fed back the servo output back to the input. Even though I used a 500/1 divider to reduce noise injection, I, unfortunately got a bad batch of LF411 fet input op amps that added excessive low frequency noise to the input. The measurements of the Vendetta Research noise done by 'Stereophile' appears to show this problem. This was an early unit that used the LF411, as it was about the only device available. Later, I changed to the AD711 (a Scott Wurcer upgrade to the LF411), exclusively, and removed most of the problem. However, I don't use the same servo connection anymore in my latest design, so I hope to have removed any significant 1/f contribution to the latest design. Even 500/1 is not really enough.
Secondly, I used the equivalent of a 10 ohm wire wound pot at the input that did 3 things: 1. bias control of the fets, 2. a way to 2'nd harmonic distortion cancel the slightly different Gm of the N and P parts principally due to mobility differences between electrons and holes, 3. to provide a ground return for the servo.
As the servo doesn't have to be there, anymore, then the resistors can be reduced to virtually nothing, much as SYN08 has done. The input resistors (pot) added only 2.5 ohms extra noise anyway, but improvement can be made, if the effective resistor value is reduced.
These are not easy circuits to build. They require excessive matching and noise screening to do it right. A production nightmare!
This number has been achieved before by Japanese designs back in the late '70's and early 80's using bipolar devices, but they never sounded as good as a fet design. It might be RFI sensitivity and dynamic range that makes the difference.
This topic still remains inside the Blowtorch thread, because an earlier version of this input stage is in the Blowtorch picture (11) and also is in my own personal unit.
I made two mistakes in the original design that is more than 25 years old now. One important one was where I fed back the servo output back to the input. Even though I used a 500/1 divider to reduce noise injection, I, unfortunately got a bad batch of LF411 fet input op amps that added excessive low frequency noise to the input. The measurements of the Vendetta Research noise done by 'Stereophile' appears to show this problem. This was an early unit that used the LF411, as it was about the only device available. Later, I changed to the AD711 (a Scott Wurcer upgrade to the LF411), exclusively, and removed most of the problem. However, I don't use the same servo connection anymore in my latest design, so I hope to have removed any significant 1/f contribution to the latest design. Even 500/1 is not really enough.
Secondly, I used the equivalent of a 10 ohm wire wound pot at the input that did 3 things: 1. bias control of the fets, 2. a way to 2'nd harmonic distortion cancel the slightly different Gm of the N and P parts principally due to mobility differences between electrons and holes, 3. to provide a ground return for the servo.
As the servo doesn't have to be there, anymore, then the resistors can be reduced to virtually nothing, much as SYN08 has done. The input resistors (pot) added only 2.5 ohms extra noise anyway, but improvement can be made, if the effective resistor value is reduced.
These are not easy circuits to build. They require excessive matching and noise screening to do it right. A production nightmare!
Sigurd Ruschkow said:Lyra's cartridges have about 5 Ohms DCR.
Still, the LP's noise is around 60-70 dB, so actually no need to make a 80 dB MC RIAA amp.
However, it is also the challenge to make super low noise amps that are part of the reason to make one
Sigurd
It does make sense to have a phono stage with noise floor lower than the source, since the stage's noise is being added to the source.
- Status
- Not open for further replies.