A thread late last year regarding the Pacific passive RIAA preamp inspired me to try some simple open loop circuits to replace my ancient Nikko Beta II preamp. The constraint I set was to use no global feedback. The gain cells I used are shown in the attached file.
The circuit on the left shows the gain cell used in the phono preamp. Like the Pacific preamp, it is a simple JFET common source amplifier, but with the addition of a source follower buffer to supply some output drive capability. The addition of the buffer also makes calculating the RIAA network much easier, as without the buffer you have to add in the contribution of the drain resistor in parallel with the output impedance of the JFET at the drain (no, it's not a perfect current source...). PN4393 JFETS were used in the buffer, as their high IDSS allows relatively high output current. I also had a bunch of them....
I aimed for a first stage gain of 40, and a second stage gain of 30, which gives a 1kHz gain of about 40dB when the RIAA network is factored in. The first stage had no source resistor. One was added to the second stage to reduce overall gain and to increase the overload capability of the second stage. I juggled values to get both the proper gain and approximately 15V at the drain of the 2SK170 for maximum symmetric output swing. Simulation of the gain cell using Orcad showed about 0.2% distortion, flat across the audio range, mostly 2nd harmonic. Measurements with an Audio Precision analyzer showed about 0.4% distortion, mostly 2nd harmonic.
For the line amp stage, I started with a circuit similar to the RIAA gain stage, but as no global feedback was used, it was difficult to tailor the gain and properly center the output voltage to 15V. I used all PN4393s for the line amp, and ended up with a stage gain of about 16 (I would have liked a gain of 5). A gain of 16 was way too much, and made volume adjustment very touchy. I ended up with a unity gain buffer circuit as shown in the right-hand drawing. Use of P-channel JFETs in the input stage provided some distortion cancellation according to simulation (0.02% vs. 0.2%). A nice feature of the new line amp is that the preamp is now non-inverting. No Audio Precision measurements were made, as I made the circuit changes late last night, and the analyzer is at work.
I used a pair of Bourns 100k linear pots for volume/balance controls (either cermet or plastic, I'm not sure which), and law-faked them with a pair of Dale 19.2k MF resistors. I'll try other pots as the mood hits me.
So far, the overall impresssion is thumbs up, though I probably will need to improve or replace my speakers before I can get the full measure of this beast. It's been on for less than 24 hours, so I still have a lot of auditioning to do. I plan later down the line to sub in some folded cascode amplifiers with global feedback and try to see if I can tell the difference.
The circuit on the left shows the gain cell used in the phono preamp. Like the Pacific preamp, it is a simple JFET common source amplifier, but with the addition of a source follower buffer to supply some output drive capability. The addition of the buffer also makes calculating the RIAA network much easier, as without the buffer you have to add in the contribution of the drain resistor in parallel with the output impedance of the JFET at the drain (no, it's not a perfect current source...). PN4393 JFETS were used in the buffer, as their high IDSS allows relatively high output current. I also had a bunch of them....
I aimed for a first stage gain of 40, and a second stage gain of 30, which gives a 1kHz gain of about 40dB when the RIAA network is factored in. The first stage had no source resistor. One was added to the second stage to reduce overall gain and to increase the overload capability of the second stage. I juggled values to get both the proper gain and approximately 15V at the drain of the 2SK170 for maximum symmetric output swing. Simulation of the gain cell using Orcad showed about 0.2% distortion, flat across the audio range, mostly 2nd harmonic. Measurements with an Audio Precision analyzer showed about 0.4% distortion, mostly 2nd harmonic.
For the line amp stage, I started with a circuit similar to the RIAA gain stage, but as no global feedback was used, it was difficult to tailor the gain and properly center the output voltage to 15V. I used all PN4393s for the line amp, and ended up with a stage gain of about 16 (I would have liked a gain of 5). A gain of 16 was way too much, and made volume adjustment very touchy. I ended up with a unity gain buffer circuit as shown in the right-hand drawing. Use of P-channel JFETs in the input stage provided some distortion cancellation according to simulation (0.02% vs. 0.2%). A nice feature of the new line amp is that the preamp is now non-inverting. No Audio Precision measurements were made, as I made the circuit changes late last night, and the analyzer is at work.
I used a pair of Bourns 100k linear pots for volume/balance controls (either cermet or plastic, I'm not sure which), and law-faked them with a pair of Dale 19.2k MF resistors. I'll try other pots as the mood hits me.
So far, the overall impresssion is thumbs up, though I probably will need to improve or replace my speakers before I can get the full measure of this beast. It's been on for less than 24 hours, so I still have a lot of auditioning to do. I plan later down the line to sub in some folded cascode amplifiers with global feedback and try to see if I can tell the difference.
Attachments
After a bit more listening, what I can say is that the new preamp seems to bring me closer to the source material than the old Nikko preamp, which had about 5 electrolytics in the signal path, as well as some regrettable design shortcuts. If the source material has been lovingly recorded, the results can be amazing. Two recordings in particular stand out. On vinyl, it's the Muhal Richard Abrams Octet "View from Within" on Black Saint (LP). The instruments are picked out with razor sharpness, and you feel like you're right there with the performers. For CD, it's the Berlin Philharmonic conducted by von Karajan with Anne-Sophie Mutter on violin, performing Beethoven's violin concerto. I've used this recording as a standard for a while, and each improvement I make in the system reveals more.
I selected FETs for the phono preamp gain cells to get decent drain voltage centering, especially for the second stage. The second stage had a source degeneration resistor, which would help to equalize gain somewhat. The gain between both channels of the preamp was pretty well matched, according to scans I did with an HP4194 gain-phase analyzer I have at work. I ended up using matched GR devices in the first stage and matched BL devices for the second stage. The next time I do this, I'll use matched BL devices with degeneration for both stages.
I wasn't aware of the earlier thread, but it covers some paths I've also troddenindependently. When I saw the modified Pacific design in Mad_K's thread late in 2003, one of the first things I thought about was to include cascoding, and I laid out my PCB to include cascode JFETs. I'm also aware of the cascoding used on the Pearl preamp. I used a JFET instead of a bipolar for the cascode element, and chose the cascode FET characteristics such that there would be 4-5V on the drain of the bottom JFET. The resulting gain cell is shown in the attached figure. I later removed the cascode FETs, because simulation showed that they increased odd order distortion at high output levels. I might try adding them in again to see if I can hear any difference. However, I'll wait on this until I have a chance to live for a while with my present arrangement.
Output buffers were included on each stage for RIAA accuracy and to reduce the loading on the gain JFET. I originally used a 2SK170GR for each stage with no source degeneration. The overall gain of the preamp was way too large for the cartridge I was using (Grado gold), so the second stage was replaced with a 2SK70BL with a source resistor. I also had some concerns about driving the gate of the second stage JFET gate positive during transients. The source degeneration resistor provides some back bias to the JFET gate to make this less likely to happen. The overall preamp is shown in the figure. I should add that one of the constraints in my design philosophy for this project was to use all JFETs as well as no global feedback, so the buffers used between stages were source followers with JFET current source loading.
I've been doing some thinking about RIAA accuracy. Using an unbuffered and non-cascode common souce amplifier feeding a passive amplifier, one needs to adjust the value of the first resistor in the RIAA network to include the parallel combination of the drain resistor of the amplifier and the drain impedance of the JFET. There is a problem with this in that the drain impedance of the JFET is not constant, but shifts with bias current (or as you traversea load line). You can see this immediately if you look at the V-I curves in the 2SK170 data sheet. Using a cascode transistor (or FET) will definitely dramatically increase the output impedance of the JFET( as well as making it more constant) by pinning its drain at a constant voltage. This may explain the improvement in listening characteristics observed by KYW due to cascoding.
I elected to use followers instead of the cascode to isolate the RIAA network from the gain FET. This eliminates the effect of JFET output impedance shift, and makes it much easier to choose the RIAA network values as well.
This afternoon, for kicks, I did a bunch of PSPICE simulations for a JFET sommon source amplifier using a 2SK170 JFET. I added cascode and/or an output follwer to look at their effect on the output distortion magnitude and distribution. I adjusted the values of source and drain resistor to get a stage gain of about 40, and ran the simulation with low (1mV) and high (40mV) excitation levels to see the difference in output distortion. A common source amplifier with cascode JFET (PN4393) driven at low level has about 0.013% distortion at 10kHz, overwhelmingly 2nd harmonic. The same circuit with the same excitation level with a source follower output has about 0.015% distortion, with just a touch more 3rd harmonic, all other harmonics at insignificant levels. The source follower doesn't appear to add much distortion on its own.
Interesting things happened at 40mV drive levels. A common source amplifier with follower and no cascode has 0.4% distortion at 10kHz, with a ratio of about 10:1 between 2nd and 3rd harmonics, and no significant higher order distortion. I observed very similar measurements while testing my this type of gain cell circuit using an Audio Precision analyzer.
When a cascode circuit is added with no other circuit changes the distortion goes up to 0.5% at 40mV drive. This is not too bad, except the 2nd and 3rd harmonic are now almost equal in amplitude. It will be interesting to see if I can verify this result with actual measurements
Output buffers were included on each stage for RIAA accuracy and to reduce the loading on the gain JFET. I originally used a 2SK170GR for each stage with no source degeneration. The overall gain of the preamp was way too large for the cartridge I was using (Grado gold), so the second stage was replaced with a 2SK70BL with a source resistor. I also had some concerns about driving the gate of the second stage JFET gate positive during transients. The source degeneration resistor provides some back bias to the JFET gate to make this less likely to happen. The overall preamp is shown in the figure. I should add that one of the constraints in my design philosophy for this project was to use all JFETs as well as no global feedback, so the buffers used between stages were source followers with JFET current source loading.
I've been doing some thinking about RIAA accuracy. Using an unbuffered and non-cascode common souce amplifier feeding a passive amplifier, one needs to adjust the value of the first resistor in the RIAA network to include the parallel combination of the drain resistor of the amplifier and the drain impedance of the JFET. There is a problem with this in that the drain impedance of the JFET is not constant, but shifts with bias current (or as you traversea load line). You can see this immediately if you look at the V-I curves in the 2SK170 data sheet. Using a cascode transistor (or FET) will definitely dramatically increase the output impedance of the JFET( as well as making it more constant) by pinning its drain at a constant voltage. This may explain the improvement in listening characteristics observed by KYW due to cascoding.
I elected to use followers instead of the cascode to isolate the RIAA network from the gain FET. This eliminates the effect of JFET output impedance shift, and makes it much easier to choose the RIAA network values as well.
This afternoon, for kicks, I did a bunch of PSPICE simulations for a JFET sommon source amplifier using a 2SK170 JFET. I added cascode and/or an output follwer to look at their effect on the output distortion magnitude and distribution. I adjusted the values of source and drain resistor to get a stage gain of about 40, and ran the simulation with low (1mV) and high (40mV) excitation levels to see the difference in output distortion. A common source amplifier with cascode JFET (PN4393) driven at low level has about 0.013% distortion at 10kHz, overwhelmingly 2nd harmonic. The same circuit with the same excitation level with a source follower output has about 0.015% distortion, with just a touch more 3rd harmonic, all other harmonics at insignificant levels. The source follower doesn't appear to add much distortion on its own.
Interesting things happened at 40mV drive levels. A common source amplifier with follower and no cascode has 0.4% distortion at 10kHz, with a ratio of about 10:1 between 2nd and 3rd harmonics, and no significant higher order distortion. I observed very similar measurements while testing my this type of gain cell circuit using an Audio Precision analyzer.
When a cascode circuit is added with no other circuit changes the distortion goes up to 0.5% at 40mV drive. This is not too bad, except the 2nd and 3rd harmonic are now almost equal in amplitude. It will be interesting to see if I can verify this result with actual measurements
Attachments
Have you considered first stage Cgd, Miller effect and its effects on cartridge loading? I don't know what recommended capacitive loading is for Grado Gold, but most MM carts seem to use 200 - 500 pF, and if Vds = 15, Av1 = and Cgd = 5pF then effective Cgs is about 180pF. Cascoding first stage reduces Cgs to about 35pF. I would be more concerned about the non-linearity of Cgs than about the exact value, simulation with an LCR input network representing your cartridge might offer insight.
I would be more concerned with second stage distortion since the input is larger than for the first stage. If you have 40mV cartridge output the distortion from the record and cartridge is probably a lot higher than 1%.
Personally I like tubes in phono preamps, but if I were building an SS preamp I would consider something like what you or KYW have worked on.
BTW, have you considered active EQ? It's a PITA to calculate EQ values for low Avol but offers S/N advantages since 2nd stage noise is attenuated.
I would be more concerned with second stage distortion since the input is larger than for the first stage. If you have 40mV cartridge output the distortion from the record and cartridge is probably a lot higher than 1%.
Personally I like tubes in phono preamps, but if I were building an SS preamp I would consider something like what you or KYW have worked on.
BTW, have you considered active EQ? It's a PITA to calculate EQ values for low Avol but offers S/N advantages since 2nd stage noise is attenuated.
A Grado cartridge is less sensitive to the value of loading capacitance because it has about 1/10 the inductance of the average MM cartridge (45mH vs. 400mH-1H). Grado doesn't even specify a recommended loading capacitor value, they just say that the cardridge is relatively insensitive to capacitive loading. I'm using 270pF and 16k for loading right now, though this could (and probably will) change. I eventually plan to get my hands on a test record with pink noise tracks to test the whole cartridge-preamp system. This should tell me a lot about what I should be using for loading, especially in conjunction with a scope with FFT function. Old Colony sells a test record with pink noise tracks, I just have to get around to ordering it. I have a nice digital scope with FFT function so this should make testing easy.
I forgot about the contribution of the miller capacitance of the FET. Cascoding will certainly minimize the effect of this. Using a Grado cartridge also might make it relatively unimportant. Another possibility would be to use a PN4393 or J309/J310 for the first JFET. These FETS have reasonable gain and audio noise specs (though not near as nice as the 2SK170). However, their gate capacitance and reverse transfer capacitance are orders of magnitude lower than that of the 2SK170. This is no surprise, as the PN4393 was originally intended as an extremely fast chopper/analog switch, while the J309 and 310 are RF FETs. You won't find audio noise specs in the data sheets for these FETs, but if you go to the characteristic sheets for the FET family (Siliconix publishes them, for example), you see the typical noise specs for the device family, and they are pretty good. The PN4393 has been mentioned as a "trade secret" sort of low noise audio amp for quite some time. I first saw mention of it in some National app notes quite a few years back.
The J110 is another one of these trade secret devices, though the units I have on hand have an IDSS that is too high (>100ma) for this application. Throttling the drain current to a reasonable value with a big source resistor would result in too small of a stage gain, and I have no desire to use a source bypass cap to bring the gain back up. Some folks have mentioned using a few devices like this
(or even nastier ones like the J105) in parallel with a relatively small load resistor and drain supply for use as a MC booster. I haven't tried this, as the only MC cartridge I have at present is a Sumiko Blue Point, which has a high enough output to make gymnastics like this unnecessary.
Ths simulation runs I did yesterday seem to point out that it might be advantageous to cascode or otherwise reduce the capacitance of the first stage, but to leave the second stage alone. When I have done enough listening to get a good, consistent impression of the preamp, then it will be time to do some messing around with this in mind.
I don't have any great desire to do any active feedback designs at this time, though I have some OPA627s and could crank one out any time I feel like it. I want to put my folded cascode boards in the box and test them before I even think about using opamps. I don't have any problems with perceived noise at present. The only thing objectionable coming out of the speakers is a very tiny trace of power line hum ( very tiny), probably due to the fact I used a cheap Ten-Tec box with plastic sides. When I do a more permanant setup, I'll use an all-metal enclosure. The one I'm using right now is just a cheap. plain-looking test bed so that I won't shed any tears about hacking it up to accomodate various experiments.
I forgot about the contribution of the miller capacitance of the FET. Cascoding will certainly minimize the effect of this. Using a Grado cartridge also might make it relatively unimportant. Another possibility would be to use a PN4393 or J309/J310 for the first JFET. These FETS have reasonable gain and audio noise specs (though not near as nice as the 2SK170). However, their gate capacitance and reverse transfer capacitance are orders of magnitude lower than that of the 2SK170. This is no surprise, as the PN4393 was originally intended as an extremely fast chopper/analog switch, while the J309 and 310 are RF FETs. You won't find audio noise specs in the data sheets for these FETs, but if you go to the characteristic sheets for the FET family (Siliconix publishes them, for example), you see the typical noise specs for the device family, and they are pretty good. The PN4393 has been mentioned as a "trade secret" sort of low noise audio amp for quite some time. I first saw mention of it in some National app notes quite a few years back.
The J110 is another one of these trade secret devices, though the units I have on hand have an IDSS that is too high (>100ma) for this application. Throttling the drain current to a reasonable value with a big source resistor would result in too small of a stage gain, and I have no desire to use a source bypass cap to bring the gain back up. Some folks have mentioned using a few devices like this
(or even nastier ones like the J105) in parallel with a relatively small load resistor and drain supply for use as a MC booster. I haven't tried this, as the only MC cartridge I have at present is a Sumiko Blue Point, which has a high enough output to make gymnastics like this unnecessary.
Ths simulation runs I did yesterday seem to point out that it might be advantageous to cascode or otherwise reduce the capacitance of the first stage, but to leave the second stage alone. When I have done enough listening to get a good, consistent impression of the preamp, then it will be time to do some messing around with this in mind.
I don't have any great desire to do any active feedback designs at this time, though I have some OPA627s and could crank one out any time I feel like it. I want to put my folded cascode boards in the box and test them before I even think about using opamps. I don't have any problems with perceived noise at present. The only thing objectionable coming out of the speakers is a very tiny trace of power line hum ( very tiny), probably due to the fact I used a cheap Ten-Tec box with plastic sides. When I do a more permanant setup, I'll use an all-metal enclosure. The one I'm using right now is just a cheap. plain-looking test bed so that I won't shed any tears about hacking it up to accomodate various experiments.
Be prepared for some surprises on third octave noise measurements beyond 10kHz or so. Remember the old Telarc Omnidisk? Well they at least warned you that the low frequency crackle you heard was not anything wrong (yeah, just the 10% or more IMD mistracking distortion). I never found a stylus that would track even a fresh test record. I have the whole set of RCA STR's and the Telarc, if you end up needing a loaner drop me a line.
I'm used to Shure carts, I guess the Grados have less LC resonance. Yeah, second stage Cin doesn't matter much, especially with passive EQ.
Vn of 4393s is probably fine for MM levels, 4nv/sqrt(Hz) is OK, but may have higher Vgs at 1mA or so than the 2sk170s so Rs may need to be larger and therefore noisier. You don't mention your Rs and Rd values, or Id. You might consider 2sk117, higher Vn than 170 but also less Cds, I think there is even a cross to an NTE part.
Test record is the way to go for RIAA EQ verification. I have a couple of NAB test records, with single tones at low level from 30 to 15k. Yeah, the crackle is kinda bad above 10kHz even at the low level. I built an inverse RIAA to tune my RIAA EQ to +/-0.1dB, after the test record I was happy with +/- 0.5dB.
Active EQ RIAA doesn't have to be done with opamps, it was the standard approach before SS, Dyna PAS3 and Eico HF85 are examples. Active EQ with op amps is easy to calculate for because of the high open loop gain. It's not so easy when you have only 60dB at 20Hz, but Lipshitz worked it out in the late '70s. My current tube phono is 2-stage active for all poles, I'm thinking of modifying it for passive on the 2122Hz pole, but I'm still using the test chassis and need to rebuild it into a decent one
Vn of 4393s is probably fine for MM levels, 4nv/sqrt(Hz) is OK, but may have higher Vgs at 1mA or so than the 2sk170s so Rs may need to be larger and therefore noisier. You don't mention your Rs and Rd values, or Id. You might consider 2sk117, higher Vn than 170 but also less Cds, I think there is even a cross to an NTE part.
Test record is the way to go for RIAA EQ verification. I have a couple of NAB test records, with single tones at low level from 30 to 15k. Yeah, the crackle is kinda bad above 10kHz even at the low level. I built an inverse RIAA to tune my RIAA EQ to +/-0.1dB, after the test record I was happy with +/- 0.5dB.
Active EQ RIAA doesn't have to be done with opamps, it was the standard approach before SS, Dyna PAS3 and Eico HF85 are examples. Active EQ with op amps is easy to calculate for because of the high open loop gain. It's not so easy when you have only 60dB at 20Hz, but Lipshitz worked it out in the late '70s. My current tube phono is 2-stage active for all poles, I'm thinking of modifying it for passive on the 2122Hz pole, but I'm still using the test chassis and need to rebuild it into a decent one
Scott, nice to hear from you again. I hope the people that made the test record I'm planning on buying had the sense to cut the pink noise tracks at reasonable levels. Tracking these will probably be challenging enough without making them part of the obstacle course as well. Thanks for pointing out some possible nasties.
To Nuvistor - I won't think of doing active feedback with the simple discrete preamp because of the interaction with cartridge impedances and such. This is one of the justifcations Holman cited for using a diff pair in the front of his preamp design, to get the feedback network away from the cartridge. I have considered putting the high frequency equalization around the first stage of my folded cascode preamp, as it is just a pair of discrete opamps anyway. It would be pretty simple to drop an RC across the main feedback resistor of the first stage to get the 2122Hz pole and 50kHz zero. This removes some of the objections regarding high frequency overload usually cited as a disadvantage for passive RIAA designs.
Even though the PN4393 will require a larger source resistor to keep the drain current within bounds, I don't think the extra noise will be an issue at all for MM or MI applications. Having said that, for the first stage, I would select the 4393s for low IDSS. Mouser sells Fairchild PN4393s for cheap enough to allow you to buy a small mountain of them and select to your heart's content. The extra reverse bias on the gate from a larger source resistor will also give you some more overdrive tolerance. In the case of MC amps, you just run the FETs at full current and take what you get for drain current, just like Boberly did for his MC preamp design.
To Nuvistor - I won't think of doing active feedback with the simple discrete preamp because of the interaction with cartridge impedances and such. This is one of the justifcations Holman cited for using a diff pair in the front of his preamp design, to get the feedback network away from the cartridge. I have considered putting the high frequency equalization around the first stage of my folded cascode preamp, as it is just a pair of discrete opamps anyway. It would be pretty simple to drop an RC across the main feedback resistor of the first stage to get the 2122Hz pole and 50kHz zero. This removes some of the objections regarding high frequency overload usually cited as a disadvantage for passive RIAA designs.
Even though the PN4393 will require a larger source resistor to keep the drain current within bounds, I don't think the extra noise will be an issue at all for MM or MI applications. Having said that, for the first stage, I would select the 4393s for low IDSS. Mouser sells Fairchild PN4393s for cheap enough to allow you to buy a small mountain of them and select to your heart's content. The extra reverse bias on the gate from a larger source resistor will also give you some more overdrive tolerance. In the case of MC amps, you just run the FETs at full current and take what you get for drain current, just like Boberly did for his MC preamp design.
This is nice for the "circuit bag", but I don't have an MC that needs it at this point. The circuit is one of the simpler implementations of a folded cascode I've seen.
So far, all the ballyhoo on this thread has been about the RIAA pramp section, but the buffer ciruit has been doing yeoman service with my CDs and tuner lately... There appears to be enough signal level (even at unity gain) to drive my amp to a comfortable volume. All I need to do is to get up off my behind and adjust the level control on my tuner to match those for LP and CD material, and all will be sweetness and light...
So far, all the ballyhoo on this thread has been about the RIAA pramp section, but the buffer ciruit has been doing yeoman service with my CDs and tuner lately... There appears to be enough signal level (even at unity gain) to drive my amp to a comfortable volume. All I need to do is to get up off my behind and adjust the level control on my tuner to match those for LP and CD material, and all will be sweetness and light...
Hi wrenchone -
You seem determined to proceed to a working prototype, keep us posted on your progress.
Hmm - is this the case for all active feedback implementations? Unfortunately I don't have Holman's AES paper, but I did look at his patent 4117412 where he discusses this, with the comment "This high frequency interaction problem was not significant in vacuum tube preamplifiers, due to their characteristic input capacitance and resistance". I think he's referring to the resistance between emitter and ground (Re) of a bipolar input stage being reflected to the input by hfe*(re+Re), and the Re and rin in an active RIAA pre using single-ended gain stages will decrease with frequency since the feedback network is effectively in parallel with Re, while the Re and rin in the differential bipolar input stage will be constant with frequency due to the isolation between feedback network and Re. This doesn't apply to JFETs or tubes since they don't have hfe, but Cgs or Cgd is an issue, in both passive and active RIAA.
I think it's unfortunate that active feedback RIAA "don't get no respect", I think it works well if done well, but I guess I'm in the minority here. It's certainly harder to do an active RIAA design than a passive using 2 single-ended discrete gain stages, and done well needs more parts, but it reduces second stage noise contribution, and stabilizes gain if you're using discrete gain stages. In the classic designs Avol is less than 60dB and open loop Ro is high or uncertain, I use mu followers to improve this.
I used to use passive RIAA using 2 op amps, the last I tried was with OPA627 and LM6171, sound is accurate and clean, but my tube active RIAA is generally more involving and more fun to listen to, though noise level with MM is a little higher and bass is not as well defined.
It's been about 4 months since I finished the tube RIAA, I'm itchin' to try something else....
You seem determined to proceed to a working prototype, keep us posted on your progress.
Hmm - is this the case for all active feedback implementations? Unfortunately I don't have Holman's AES paper, but I did look at his patent 4117412 where he discusses this, with the comment "This high frequency interaction problem was not significant in vacuum tube preamplifiers, due to their characteristic input capacitance and resistance". I think he's referring to the resistance between emitter and ground (Re) of a bipolar input stage being reflected to the input by hfe*(re+Re), and the Re and rin in an active RIAA pre using single-ended gain stages will decrease with frequency since the feedback network is effectively in parallel with Re, while the Re and rin in the differential bipolar input stage will be constant with frequency due to the isolation between feedback network and Re. This doesn't apply to JFETs or tubes since they don't have hfe, but Cgs or Cgd is an issue, in both passive and active RIAA.
I think it's unfortunate that active feedback RIAA "don't get no respect", I think it works well if done well, but I guess I'm in the minority here. It's certainly harder to do an active RIAA design than a passive using 2 single-ended discrete gain stages, and done well needs more parts, but it reduces second stage noise contribution, and stabilizes gain if you're using discrete gain stages. In the classic designs Avol is less than 60dB and open loop Ro is high or uncertain, I use mu followers to improve this.
I used to use passive RIAA using 2 op amps, the last I tried was with OPA627 and LM6171, sound is accurate and clean, but my tube active RIAA is generally more involving and more fun to listen to, though noise level with MM is a little higher and bass is not as well defined.
It's been about 4 months since I finished the tube RIAA, I'm itchin' to try something else....
To Nuvistor - a single ended preamp with active feedback will work with everything that is hanging off the base/gate/grid, including the cartridge. This makes a case for using a diff pair to separate the feedback network from the cartridge. I'll still wait until I try out my folded cascode preamp to try the active feedback on the front end.
To Werner - The single ended folded cascode front end you cited has some possibilities for working with high IDSS FETs because you can divide up the currents to run a high bias current in the JFET and a more modest current in the cascode transistor to better center the voltage on the load resistor. However, the simple implementation with a resistor on top of the JFET drain will have pretty poor ripple rejection.
I show in the figure a different version using a current source on top of the JFET. This will give better line rejection. I've used a variation of this circuit for a charge sensitive amplifier for nuclear instrumentation with very good results. However, the simulation of this circuit shows a disturbing characteristic. Though the total THD is fairly low (0.044%), the distribution of harmonics is skewed toward odd order products (2nd- 6.6E-5, 3rd- 2E-4, 4th- 6.1E-5, 5th- 2.4E-4). A simple common source or conventional cascode front end simulate and measure (in the amp I've actually built) as having a much more benign harmonic distribution, dominated by second order for the simple amplifier and 2nd/3rd for the conventional cascode.
To Werner - The single ended folded cascode front end you cited has some possibilities for working with high IDSS FETs because you can divide up the currents to run a high bias current in the JFET and a more modest current in the cascode transistor to better center the voltage on the load resistor. However, the simple implementation with a resistor on top of the JFET drain will have pretty poor ripple rejection.
I show in the figure a different version using a current source on top of the JFET. This will give better line rejection. I've used a variation of this circuit for a charge sensitive amplifier for nuclear instrumentation with very good results. However, the simulation of this circuit shows a disturbing characteristic. Though the total THD is fairly low (0.044%), the distribution of harmonics is skewed toward odd order products (2nd- 6.6E-5, 3rd- 2E-4, 4th- 6.1E-5, 5th- 2.4E-4). A simple common source or conventional cascode front end simulate and measure (in the amp I've actually built) as having a much more benign harmonic distribution, dominated by second order for the simple amplifier and 2nd/3rd for the conventional cascode.
Attachments
Out of a feeling of needing to look at everything with an unjaundiced eye, I did a simulation of a single ended folded cascode circuit using a resistor at the drain. I took the circuit shown in my previous post, replaced the current source at the drain with a single resistor, and eliminated the voltage source between the base of the cascode transistor and +30V. The output buffer and and all other values remained the same.
I sized the voltage divider on the cascode transistor to provide 5V from the base to +30V, and pegged the resistor at the drain of the input JFET so that the FET had about 8ma of quiescent current and the cascode transistor had about 3ma. This took 360 ohms from the input FET drain to +30V.
Simulation gave some surprising results. The gai for the circuit as I have described it was almost exactly 40. THD was about 0.004% at 10kHz with 1mV drive and 30k load, the bulk being 2nd order, with the other components vanishingly small. It will be interesting if the circuit measures as well as it simulates, though I've had pretty good correlation between the stuff I've been able to measure and my simulation results. I've been running my simulations with a rel tol of 0.0001 and 20nsec time steps.
The one big fly in the ointment is still that the circuit will have poor line rejection, as there is a direct path from 30V down through the cascode transistor to the load resistor. This may not be an issue for people willing to use a really good regulator or a battery supply. I will have to do some thinking to figure out why this circuit measures (or at least, simulates) so much better than the option using a current source. Maybe others can offer some insight.
I sized the voltage divider on the cascode transistor to provide 5V from the base to +30V, and pegged the resistor at the drain of the input JFET so that the FET had about 8ma of quiescent current and the cascode transistor had about 3ma. This took 360 ohms from the input FET drain to +30V.
Simulation gave some surprising results. The gai for the circuit as I have described it was almost exactly 40. THD was about 0.004% at 10kHz with 1mV drive and 30k load, the bulk being 2nd order, with the other components vanishingly small. It will be interesting if the circuit measures as well as it simulates, though I've had pretty good correlation between the stuff I've been able to measure and my simulation results. I've been running my simulations with a rel tol of 0.0001 and 20nsec time steps.
The one big fly in the ointment is still that the circuit will have poor line rejection, as there is a direct path from 30V down through the cascode transistor to the load resistor. This may not be an issue for people willing to use a really good regulator or a battery supply. I will have to do some thinking to figure out why this circuit measures (or at least, simulates) so much better than the option using a current source. Maybe others can offer some insight.
wrenchone said:
Hehe. I told you so, obliquely, in my reference to the Vendetta. Or did you think that John Curl would not know about putting a politically-correct current source on top of the cascode? If a measly resistor is good enough for JC, then it is good enough for me. This said, I am working on a breadboard that can accomodate resistor as well as current source, and a variety of cascode transistor biasing schemes. These are bound to be important as well. The other JC once wrote here that he uses an individual stabilised voltage for each and every cascode Q.
Ripple rejection? That's tackled elsewhere. One needs a very very good power supply for an MC frontend anyway. Luckily it is rather easy to make a good voltage regulator when its load is only a folded cascode.
I seem to remember that my circuit measured -70dB for the second harmonic when fed with 1mV. Don't remember whether this was at 300 or 1000Hz.
As an aside: what JFET models are you using? Specifically for the Japanese ones (170, 74, 369, 389, ...)?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.