This thread presents Curly-F4, a folded cascode gain stage for the Pass F4 amplifier, built on a small, piggy-backed PCB on top of the DIYAudio Store Cviller F4 PCB.
The challenge of implementing a preamp for the F4 is to achieve an output voltage swing of 20 volts peak, a gain of at least 6X, low distortion, and low noise. Some of the previous gain stages that are suitable for the F4 are:
The challenge in achieving a 20V output swing with J74 and K170 input JFETs is obeying the gate-drain voltage and power dissipation safe operating area, limited by the J74 with Vgdmax=25V and Pdmax=400mW. The solution is to set the cascode node voltage to 23V and use JFETs with Idds in the range of 6-7mA, thus limiting the power dissipation to around 160mW.
BC550C and BC560C BJTs were chosen for the cascode devices because 1) they have the needed voltage (43V max) and power power dissipation (180mW avg), and 2) they appeared to perform better than MOSFETs for the task. No P-channel JFET can handle the 43V requirement.
The power supply voltages of +50V and -50V were chosen because 1) I had a suitable power transformer, and 2) at least 30 volts is needed for supplying the (nearly) constant cascode current. I experimented with lower voltage supplies using DN2540 based constant current sources, but determined that at least 30 volts was required for a good CCS using the DN2540. I ended up replacing the CCS with a simple resistor and using higher voltages. At the 15mA folded cascode currents, there is no significant cost for using higher voltage supplies since the +/-23 volt cascode node voltages determine the voltage requirements for the active devices.
Shown below is the schematic for the Curly Gain Stage as built. As built, the voltage gain is 10X, can be modified by changing R56 and/or R57. The 10R P3 adjustment trimmer alters the second harmonic of the preamp. Currently I have omitted resistors R62 and R63, but I would suggest using something like R62=R63=5R, and P3=100R to give finer adjustment of H2 and less susceptibility to wiper "noise". For other power supply voltages the cascode load resistors R54 and R55 should be (Vpwr-23)/15mA.
The challenge of implementing a preamp for the F4 is to achieve an output voltage swing of 20 volts peak, a gain of at least 6X, low distortion, and low noise. Some of the previous gain stages that are suitable for the F4 are:
- BA3GS: Non-feedback F5-lite. http://www.diyaudio.com/forums/pass-labs/195303-burning-amp-ba-3-a.html
- Pumpkin Preamp: Balanced UGS. http://www.diyaudio.com/forums/pass-labs/103130-pumpkin-preamp-perfect-f4.html
- Impasse Preamp: Very nice vacuum tube balanced gain stage. http://www.diyaudio.com/forums/pass-labs/136835-impasse-preamplifier.html
- LSK pre BAF 2013: Folded cascode. http://www.diyaudio.com/forums/pass-labs/244106-lsk-pre-baf-2013-a.html
- John Curl's Blowtorch: Balanced folded cascode. http://www.diyaudio.com/forums/solid-state/71189-john-curls-blowtorch-preamplifier.html
The challenge in achieving a 20V output swing with J74 and K170 input JFETs is obeying the gate-drain voltage and power dissipation safe operating area, limited by the J74 with Vgdmax=25V and Pdmax=400mW. The solution is to set the cascode node voltage to 23V and use JFETs with Idds in the range of 6-7mA, thus limiting the power dissipation to around 160mW.
BC550C and BC560C BJTs were chosen for the cascode devices because 1) they have the needed voltage (43V max) and power power dissipation (180mW avg), and 2) they appeared to perform better than MOSFETs for the task. No P-channel JFET can handle the 43V requirement.
The power supply voltages of +50V and -50V were chosen because 1) I had a suitable power transformer, and 2) at least 30 volts is needed for supplying the (nearly) constant cascode current. I experimented with lower voltage supplies using DN2540 based constant current sources, but determined that at least 30 volts was required for a good CCS using the DN2540. I ended up replacing the CCS with a simple resistor and using higher voltages. At the 15mA folded cascode currents, there is no significant cost for using higher voltage supplies since the +/-23 volt cascode node voltages determine the voltage requirements for the active devices.
Shown below is the schematic for the Curly Gain Stage as built. As built, the voltage gain is 10X, can be modified by changing R56 and/or R57. The 10R P3 adjustment trimmer alters the second harmonic of the preamp. Currently I have omitted resistors R62 and R63, but I would suggest using something like R62=R63=5R, and P3=100R to give finer adjustment of H2 and less susceptibility to wiper "noise". For other power supply voltages the cascode load resistors R54 and R55 should be (Vpwr-23)/15mA.
Attachments
Last edited:
Curly-F4 schematic and measurements
Shown below are the full schematic for Curly and my F4 implementation, distortion sweeps, and harmonic spectrum.
The F4 was biased at 1.3A per channel quiescent. Curly P3 was adjusted for slightly higher H2 vs H3 at 1 Watt. As can be seen the left and right channel sweeps are nearly identical, and THD varies only slightly until the highest frequencies.
The harmonic spectrum shows the absence of harmonics above H3, and 60Hz and its harmonics about 90dB below 1 Watt (4V peak output into 8 Ohms). Otherwise, the noise floor is very low. AC noise measurement of the output voltage with a silent input (connected to a Buffalo-IISE DAC) is about 100uV, limited by the sensitivity of the meter, and consistent with the noise spectrum.
Shown below are the full schematic for Curly and my F4 implementation, distortion sweeps, and harmonic spectrum.
The F4 was biased at 1.3A per channel quiescent. Curly P3 was adjusted for slightly higher H2 vs H3 at 1 Watt. As can be seen the left and right channel sweeps are nearly identical, and THD varies only slightly until the highest frequencies.
The harmonic spectrum shows the absence of harmonics above H3, and 60Hz and its harmonics about 90dB below 1 Watt (4V peak output into 8 Ohms). Otherwise, the noise floor is very low. AC noise measurement of the output voltage with a silent input (connected to a Buffalo-IISE DAC) is about 100uV, limited by the sensitivity of the meter, and consistent with the noise spectrum.
Attachments
Last edited by a moderator:
Curly PCB
The Curly Gain Stage was built using a PCB designed for a similar CCS based folded cascode gain stage http://www.diyaudio.com/forums/pass-labs/97540-f4-power-amplifier-385.html#post3684975. That PCB was adequate for the task, while not optimal, and is a throwaway.
The Curly PCB is attached to the F4 PCB using 3/16" hex #4 9/32" long standoffs (Mouser 761-2053T-440-AL-7). Connections are made for Gnd, F4In, F4Out, and one additional support connection at the other end of the PCB.
The following minor modifications to the Cviller F4 board are required:
Drill the Out, Gnd, and In connection holes to accept a #4 screw.
Drill and additional hole for a #4 screw between D4 and R13 as shown in the photo.
(Optional) To allow "normal" F4 operation, drill 2 holes for a 2-pin connector (Mouser 798-DF1B-2P-2.5DSA01), and wire to Gnd, and In on the F4 board. Note that the connector is incorrectly positioned in the photo below. It should be further to the left so that it can be used when the Curly board is in place.
Curly can be bypassed by removing the card, or removing the #4 screw at the F4In connector if the thru hole for the F4In is modified to disable the connection from top to bottom.
The Curly Gain Stage was built using a PCB designed for a similar CCS based folded cascode gain stage http://www.diyaudio.com/forums/pass-labs/97540-f4-power-amplifier-385.html#post3684975. That PCB was adequate for the task, while not optimal, and is a throwaway.
The Curly PCB is attached to the F4 PCB using 3/16" hex #4 9/32" long standoffs (Mouser 761-2053T-440-AL-7). Connections are made for Gnd, F4In, F4Out, and one additional support connection at the other end of the PCB.
The following minor modifications to the Cviller F4 board are required:
Drill the Out, Gnd, and In connection holes to accept a #4 screw.
Drill and additional hole for a #4 screw between D4 and R13 as shown in the photo.
(Optional) To allow "normal" F4 operation, drill 2 holes for a 2-pin connector (Mouser 798-DF1B-2P-2.5DSA01), and wire to Gnd, and In on the F4 board. Note that the connector is incorrectly positioned in the photo below. It should be further to the left so that it can be used when the Curly board is in place.
Curly can be bypassed by removing the card, or removing the #4 screw at the F4In connector if the thru hole for the F4In is modified to disable the connection from top to bottom.
Attachments
DIYAudio Store 4U Heatsink Modifications
I also prefer to attach my FETs to the heatsinks using 3-pin Molex-style connectors to make it easy to remove the PCB for modifications or to allow for testing different output FETs. This requires that the FETs are raised 1/4" using an aluminum plate which is attached to the heatsink with eight #6 screws and thermal grease. Holes are drilled for both TO-3P and TO-220 FETs. Due to availability, my F4 build actualy uses FQP12P20 and FQP19N20C TO-220 MOSFETs rather than the FQA12P20 and FQA19N20C MOSFETs shown in the schematic.
I also prefer to attach my FETs to the heatsinks using 3-pin Molex-style connectors to make it easy to remove the PCB for modifications or to allow for testing different output FETs. This requires that the FETs are raised 1/4" using an aluminum plate which is attached to the heatsink with eight #6 screws and thermal grease. Holes are drilled for both TO-3P and TO-220 FETs. Due to availability, my F4 build actualy uses FQP12P20 and FQP19N20C TO-220 MOSFETs rather than the FQA12P20 and FQA19N20C MOSFETs shown in the schematic.
Attachments
DIYAudio Store 4U Chassis Modifications
To facilitate experimentation, I made a number of modifications to the Store 4U Chassis.
I was disappointed that the heatsink attachment of the Store chassis was so difficult after full assembly. Based on my experience with a Hafler 220, I drilled 4 holes in the corners of each heatsink and in corresponding locations in the attachment brackets to allow heatsink attachment using #6 socket head screws from the OUTSIDE of the chassis. This allows the heatsink to be easily removed by disconnecting the amplifier PCB connections and removing 4 screws, which is accomplished in a couple of minutes.
To facilitate experimentation, I made a number of modifications to the Store 4U Chassis.
I was disappointed that the heatsink attachment of the Store chassis was so difficult after full assembly. Based on my experience with a Hafler 220, I drilled 4 holes in the corners of each heatsink and in corresponding locations in the attachment brackets to allow heatsink attachment using #6 socket head screws from the OUTSIDE of the chassis. This allows the heatsink to be easily removed by disconnecting the amplifier PCB connections and removing 4 screws, which is accomplished in a couple of minutes.
Attachments
Last edited:
A balanced version should be simple based on the Blowtorch topology. I am not sure about integration with a balanced amp based on the Store Cviller board. Can the X node safely cross between the two heatsinks without noise problems, or must the balanced Curly be be on a separate, single PCB. I would much prefer a balanced F4 on a single heatsink, but that involves a lot of heat in a single sink, lower PS voltages (like EUVL's F5X), or fan cooling.
Dual 8" 10.08 profile from HeatsinkUSA per side would take the heat at bias up to 2.4A with 22V rails. Lot of heat, but difference between cumulative heat of all fets(majority of heat is external to get) vs heatsink fighting to get rid of high internal heat of fets. Already have output stage designed. Dropped it to finish other things.
Buzz: You handle the heat, I can supply the gain stage!
My Curllian-F4 PCB is actually fabricated as a pair of channels that have interconnects for balanced operation. The entire pair can be mounted (somewhere) and fed to the balanced F4 pairs, or can be split, with wire (or pair) interconnecting for the "X-node".
My Curllian-F4 PCB is actually fabricated as a pair of channels that have interconnects for balanced operation. The entire pair can be mounted (somewhere) and fed to the balanced F4 pairs, or can be split, with wire (or pair) interconnecting for the "X-node".
It sounds great, and has NO audible noise with the ear against the speaker.
So far, my listening tests indicate no problems. I prefer the Impasse to the BA3GS, but I have not A/B tested the Curly4 vs. the Impasse. I have another, uncrippled F4, an Impasse, and a BA3GS. For A/B testing, all I need is switches between the amps and the speakers, and some way to equalize the gains.
They tie to the ends of the trimmer pot, ie. the JFET source pins. I am not sure what resistance would be best.
For the truth's sake, the LSK pre does produce more than that. It's just a matter of few different PS and resistor values - look at the post #25 of the said thread (higher gain => tighter bandwidth and vice versa)...
Otherwise, nice build
I stand corrected. With PS voltage and resistor values adjusted, there essentially no difference between my Curly gain stage and your LSK pre (without the output buffer). I need do look more carefully at the end-to-end square wave response to see if your use of an RC snubber might be useful.For the truth's sake, the LSK pre does produce more than that. It's just a matter of few different PS and resistor values - look at the post #25 of the said thread (higher gain => tighter bandwidth and vice versa)...
Otherwise, nice build
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.