Edmond Stuart said:
REference is Bilotti JSSC Dec 75, 516-524
I didn't read the article but read a summary in Willy Sansen book from KU Leuven.
The power of the noise current at the output of the mirror decreases inversly proportionnal to the resistor value for values of R greater than 1/gm. The 1/f noise will be lower too.
This comes from the feedback in the mirror . The feedback factor is gmR and the power decreases as the square of this factor.
R should then be greater than 1/gm.
JPV
Noise; Willy Sansen
Thank you, Jean-Pierre!
For those interested in a handsome summary of various noise sources:
rfic.fudan.edu.cn/courses/analog/Handout/Chapter04.pdf
Regards,
Edmond.
JPV said:
Thank you, Jean-Pierre!
For those interested in a handsome summary of various noise sources:
rfic.fudan.edu.cn/courses/analog/Handout/Chapter04.pdf
Regards,
Edmond.
syn08 said:
Two remarks/question from a non specialist:
You could use a beta helper in each mirror to reduce influence of mismatch in mirrors ( if this is important). Sansen shows that the injected noise from the beta helper and its resistor is insignificant because it has to be divided by beta squared again because of feedback.
Is the input resistor necessary for small signal or only for bias of the sensor?
JPV
Re: Noise; Willy Sansen
Thanks for the link Edmond. Actually, there's about 15 chapters there, heavily biased (pun intended) to opamps and such, but some may be of interest to amplifier smiths.
Jan Didden
Edmond Stuart said:
Thanks for the link Edmond. Actually, there's about 15 chapters there, heavily biased (pun intended) to opamps and such, but some may be of interest to amplifier smiths.
Jan Didden
Thanks for the link Edmond!
Great stuff, I could, however, use a little more text.
Have fun, Hannes
Great stuff, I could, however, use a little more text.
Have fun, Hannes
syn08 said:
In case you didn't realized it yet, if the impedance of the FB network in your ultra low noise pre-amp wasn't that low, it's also current feedback. 😀
PS: Hannes & Jan, you're welcome.
Hps 3.1
Ok guys, so I have listened to all the critics and I decided to redo the low noise preamp. Version is now HPS 3.1
To summarize, there were four main issues:
1. The large capacitive load that affects the slew rate and hence induces TIM.
2. Distortions are to high (although well under what a regular MC preamp is providing).
3. Not much headroom (apparently 28dB is not good enough).
4. A global servo was not available (that was my own challenge).
I have looked at these upside down. While I essentially disagree that any of the above is a really significant issue, after all it can be done better, so why not?
Unfortunately (but not unexpected) the above are conflicting with the original architecture; it is not possible to build a ultra low noise input stage (requiring very low feedback impedances) and at the same time provide (part of) the RIAA correction (requiring large capacitors, loading the input stage output). There is of course always the possibility of building a fully discrete input stage, providing as much output current for the feedback network; but one of the original goals was to use as much as possible the latest high performance opamps.
This is the reason why the input stage is now a much more simple head amp, providing a constant gain of 40 over the whole audio spectrum. This head amp has to be followed by a RIAA stage of choice; it doesn't have to be extremely low gain, and it can also be reused for MM purposes. Of course, the output stage still has to be a high voltage opamp to provide the end to end headroom.
The schematic is attached, and this is now at version HPS3.1. For output current reasons, the otherwise excellent AD797 was dropped from the list; with only 25mA of current and a max. of 50mA it limits the input stage headroom. Instead, THS4031 is now used. THS4031 has about the same noise spec as AD797, however it provides 100mA continuous output current, while the maximum output current is 150mA. It is also high speed and stable at gains>2.
Let's see how much headroom we get for an output current of 100mA:
1) U0max=I0max*Rf/2 where U0max is the max output voltage, I0max is the max output current(100mA) and Rf is the feedback resistor (39ohm)
2) The headamp gain is G=1+Rf/Rs where Rs is the source resistor (1ohm), or about G=Rf/Rs
3) By substitutin Rf in 1) we get U0max=G*I0max*Rs/2
4) But obviously G=Uomax/Uimax where Uimax is the maximum input voltage for I0max, therefore Uimax=I0max*Rs/2
Interesting enough, the maximum input voltage for a maximum output current depends only on the source resistor. In our case, Rs=1ohm, therefore the maximum input voltage is Uimax=50mV (corresponding to a maximum output voltage of 2V). The design is intended for a 0.5mV MC cartridge, so the headroom is 20*log(Uimax/0.5mV)=40dB
I have decided for an overall MC gain of 60dB @1KHz, or 40dB at 20KHz. The headamp has a gain of 40 (32dB) therefore the RIAA stage(s) have to provide a gain of 8dB (2.5) @20KHz. For a +/-28V supply and using OPA551 JFET input opamps, the output is 5Veff or about 7Vpeak.
At the other end of the spectrum, the LF gain is 80dB, therefore the RIAA stage(s) provides a gain of 48dB. The output stage maximum swing is about +/-25V, therefore the maximum LF input voltage is 2.5 mVpeak, or 1.8mVeff. This is, in combination with the input impedance (switchable from 47k to 10ohm), most likely enough to accomodate a large range of MC cartridge models, including high output MC. For a nominal input of 1.8mVeff, the headroom is still 29dB! (without pushing the THS4031 over the 100mA nominal limit). Pushing at the maximum rating, the headroom is 34dB. It can be done even better, by using a current feedback power opamp as LT1206 (250mA output current nominal, in TO220), but this is already beyond the scope and spec of this design.
On the AC/distortion front, the head amp open loop gain was increased to 60dB (by adjusting the local feedback resistor). This provides 28dB of loop gain to further reduce the distortions. Simulations are showing an ULG of 10MHz and a (lag corrected) phase shift of zero. However, the 220k resistor makes the stage very sensitive to stray capacitances, and although stability is certainly not an issue, it's probably best to wait for the final measurements to have some definitive values.
The noise is the same as in HPS3.0, around 2.5nV/rtHz (matched and sorted JFETs). Nothing has changed from this perspective.
And finally, a current servo was added. The current servo is essentially a voltage to current converter (having therefore a high output impedance, so that the gain is not affected). It currently requires three regular opamps, although it can probably done with only two. Not only does this servo help stabilizing the output, but it also relaxes a bit the Idss matching for the K170/J74 pairs (+/-2mA is acceptable, instead of +/-0.2mA). I'll be back with the full schematic in the GPP thread.
To summarize:
Gain: 60dB @1KHz
RIAA: +/-0.1dB
Noise: 0.25nV/rtHz
Max input: 1.8mVeff
Headroom: 40dB (0.5mVeff) 29dB (1.8mVeff)
Distortions: <0.01% (TBD)
I have experimented this version by cannibalizing a HPS3.0 PCB (simply not installing the RIAA correction in the input stage plus a few adjustments, including a SMD adapter instead of the AD797) and everything seems to be fine (input stage wise). I guess all the current concerns are now addressed. I'll wait for your input before running another PCB batch.
Ok guys, so I have listened to all the critics and I decided to redo the low noise preamp. Version is now HPS 3.1
To summarize, there were four main issues:
1. The large capacitive load that affects the slew rate and hence induces TIM.
2. Distortions are to high (although well under what a regular MC preamp is providing).
3. Not much headroom (apparently 28dB is not good enough).
4. A global servo was not available (that was my own challenge).
I have looked at these upside down. While I essentially disagree that any of the above is a really significant issue, after all it can be done better, so why not?
Unfortunately (but not unexpected) the above are conflicting with the original architecture; it is not possible to build a ultra low noise input stage (requiring very low feedback impedances) and at the same time provide (part of) the RIAA correction (requiring large capacitors, loading the input stage output). There is of course always the possibility of building a fully discrete input stage, providing as much output current for the feedback network; but one of the original goals was to use as much as possible the latest high performance opamps.
This is the reason why the input stage is now a much more simple head amp, providing a constant gain of 40 over the whole audio spectrum. This head amp has to be followed by a RIAA stage of choice; it doesn't have to be extremely low gain, and it can also be reused for MM purposes. Of course, the output stage still has to be a high voltage opamp to provide the end to end headroom.
The schematic is attached, and this is now at version HPS3.1. For output current reasons, the otherwise excellent AD797 was dropped from the list; with only 25mA of current and a max. of 50mA it limits the input stage headroom. Instead, THS4031 is now used. THS4031 has about the same noise spec as AD797, however it provides 100mA continuous output current, while the maximum output current is 150mA. It is also high speed and stable at gains>2.
Let's see how much headroom we get for an output current of 100mA:
1) U0max=I0max*Rf/2 where U0max is the max output voltage, I0max is the max output current(100mA) and Rf is the feedback resistor (39ohm)
2) The headamp gain is G=1+Rf/Rs where Rs is the source resistor (1ohm), or about G=Rf/Rs
3) By substitutin Rf in 1) we get U0max=G*I0max*Rs/2
4) But obviously G=Uomax/Uimax where Uimax is the maximum input voltage for I0max, therefore Uimax=I0max*Rs/2
Interesting enough, the maximum input voltage for a maximum output current depends only on the source resistor. In our case, Rs=1ohm, therefore the maximum input voltage is Uimax=50mV (corresponding to a maximum output voltage of 2V). The design is intended for a 0.5mV MC cartridge, so the headroom is 20*log(Uimax/0.5mV)=40dB
I have decided for an overall MC gain of 60dB @1KHz, or 40dB at 20KHz. The headamp has a gain of 40 (32dB) therefore the RIAA stage(s) have to provide a gain of 8dB (2.5) @20KHz. For a +/-28V supply and using OPA551 JFET input opamps, the output is 5Veff or about 7Vpeak.
At the other end of the spectrum, the LF gain is 80dB, therefore the RIAA stage(s) provides a gain of 48dB. The output stage maximum swing is about +/-25V, therefore the maximum LF input voltage is 2.5 mVpeak, or 1.8mVeff. This is, in combination with the input impedance (switchable from 47k to 10ohm), most likely enough to accomodate a large range of MC cartridge models, including high output MC. For a nominal input of 1.8mVeff, the headroom is still 29dB! (without pushing the THS4031 over the 100mA nominal limit). Pushing at the maximum rating, the headroom is 34dB. It can be done even better, by using a current feedback power opamp as LT1206 (250mA output current nominal, in TO220), but this is already beyond the scope and spec of this design.
On the AC/distortion front, the head amp open loop gain was increased to 60dB (by adjusting the local feedback resistor). This provides 28dB of loop gain to further reduce the distortions. Simulations are showing an ULG of 10MHz and a (lag corrected) phase shift of zero. However, the 220k resistor makes the stage very sensitive to stray capacitances, and although stability is certainly not an issue, it's probably best to wait for the final measurements to have some definitive values.
The noise is the same as in HPS3.0, around 2.5nV/rtHz (matched and sorted JFETs). Nothing has changed from this perspective.
And finally, a current servo was added. The current servo is essentially a voltage to current converter (having therefore a high output impedance, so that the gain is not affected). It currently requires three regular opamps, although it can probably done with only two. Not only does this servo help stabilizing the output, but it also relaxes a bit the Idss matching for the K170/J74 pairs (+/-2mA is acceptable, instead of +/-0.2mA). I'll be back with the full schematic in the GPP thread.
To summarize:
Gain: 60dB @1KHz
RIAA: +/-0.1dB
Noise: 0.25nV/rtHz
Max input: 1.8mVeff
Headroom: 40dB (0.5mVeff) 29dB (1.8mVeff)
Distortions: <0.01% (TBD)
I have experimented this version by cannibalizing a HPS3.0 PCB (simply not installing the RIAA correction in the input stage plus a few adjustments, including a SMD adapter instead of the AD797) and everything seems to be fine (input stage wise). I guess all the current concerns are now addressed. I'll wait for your input before running another PCB batch.
CG said:
Edit: +/-16V. Output swing is max 2V, for 100mA output current from the THS4031 opamp.
Closer, Syno8, but you can do even better with a mild adjustment of your feedback configuration.
CG said:
1W resistors will do just fine. The whole thing is actually a power hog, low power was not design criteria. The current mirrors, the parallel regulators and the 317/337 regulators will need onboard heatsinks. Total power dissipation is estimated at 8W average.
- Status
- Not open for further replies.