Thanks!
2sk170 shoud be fine.
There is a Rx and Cx calculator at Rod elliot site for this topology. Rx=2Ry
Linkwitz-Riley Electronic Crossover
2sk170 shoud be fine.
There is a Rx and Cx calculator at Rod elliot site for this topology. Rx=2Ry
Linkwitz-Riley Electronic Crossover
Hi,
alternative transistors, not as costly as the Toshibas SK170 and much easier to source could be the BF256C and the BF246A/247A. The former is a bit higher in noise (still sufficient for highlevel circuits), the latter features much higher Idss which allows for larger source resistor values and for higher bias currents, which could be advantageous in output buffers and buffers that need to drive capacitive loads -which is a very common case with filters.
The probabely easiest way to simulate is to download FilterPro software from the TI/BB Website. Just use Sallen-Key as Filter structure. Sim is done with the unity-gain-SallenKey structure, which is another wording for active Buffer filters ;-) Filter responses are visualized and You may choose filter charactreristics off of Bessel and Butterworth. The associated Description tutorial is SBFA001 of Nov.2001 "FilterPro MFB and Sallen-Key Low-Pass Filter Design Program".
Or the tutorial-like Filter designer http://focus.ti.com/docs/toolsw/folders/print/filter-designer.html
jauu
Calvin
alternative transistors, not as costly as the Toshibas SK170 and much easier to source could be the BF256C and the BF246A/247A. The former is a bit higher in noise (still sufficient for highlevel circuits), the latter features much higher Idss which allows for larger source resistor values and for higher bias currents, which could be advantageous in output buffers and buffers that need to drive capacitive loads -which is a very common case with filters.
The probabely easiest way to simulate is to download FilterPro software from the TI/BB Website. Just use Sallen-Key as Filter structure. Sim is done with the unity-gain-SallenKey structure, which is another wording for active Buffer filters ;-) Filter responses are visualized and You may choose filter charactreristics off of Bessel and Butterworth. The associated Description tutorial is SBFA001 of Nov.2001 "FilterPro MFB and Sallen-Key Low-Pass Filter Design Program".
Or the tutorial-like Filter designer http://focus.ti.com/docs/toolsw/folders/print/filter-designer.html
jauu
Calvin
Last edited:
Hello all,
This stuff with discreate parts is really cool.
I have used pro Rane stuff and built Rod Elliott PCB's with OpAmps active crossovers. To some degree there is added noise.
A couple of questions for you all regarding discrete parts.
What are the results, or how does it sound?
What is the noise performance of all that active circuitry? Does it hiss or hum? Does anyone have FFT outputs of the noise?
DT
All Just for fun!
This stuff with discreate parts is really cool.
I have used pro Rane stuff and built Rod Elliott PCB's with OpAmps active crossovers. To some degree there is added noise.
A couple of questions for you all regarding discrete parts.
What are the results, or how does it sound?
What is the noise performance of all that active circuitry? Does it hiss or hum? Does anyone have FFT outputs of the noise?
DT
All Just for fun!
I'll find out how my version sounds if I can get it into a box over the long weekend. I expect that biamping will drastically alter the character of my system.
The filtering is all being done at high levels - I'm surprised that noise is an issue. I'm using PN4393 jfets in my setup. They are reasonable contenders, and I haven't had any problems using them even in an RIAA preamp. No hiss or hum. For those who worry a lot, there's always the 2SK170 - it's hard to get any better than that. As for hum, that is more a problem with implementation rather than design.
The filtering is all being done at high levels - I'm surprised that noise is an issue. I'm using PN4393 jfets in my setup. They are reasonable contenders, and I haven't had any problems using them even in an RIAA preamp. No hiss or hum. For those who worry a lot, there's always the 2SK170 - it's hard to get any better than that. As for hum, that is more a problem with implementation rather than design.
Hi,
well regarding soundjudgements aren´t easy, because everybody has a different taste. To me all OP-Amp filters I built sounded artficial to a certain degree. I never doubted to listen to a reproduction of music. The first discrete JFET-filter changed that. Now there was music a lifelike quality which sounded authentic. Music instead of HiFi. I know that distortion-wise the simple Buffer is very good, but dosen´t reach a OP´s low values. But the...who cares about distortion figures, which -as we know for 30 years now- don´t correlate at all with the sonic impression? Noisewise even the BF256C settles around -100dB, the BF246A, 2SK373, 2SK170 beeing even better. So noise should be no big issue. Hum can be an issue in every circuit and typically it depends alot more on the circuits surroundings than on the circuit itself.
jauu
Calvin
well regarding soundjudgements aren´t easy, because everybody has a different taste. To me all OP-Amp filters I built sounded artficial to a certain degree. I never doubted to listen to a reproduction of music. The first discrete JFET-filter changed that. Now there was music a lifelike quality which sounded authentic. Music instead of HiFi. I know that distortion-wise the simple Buffer is very good, but dosen´t reach a OP´s low values. But the...who cares about distortion figures, which -as we know for 30 years now- don´t correlate at all with the sonic impression? Noisewise even the BF256C settles around -100dB, the BF246A, 2SK373, 2SK170 beeing even better. So noise should be no big issue. Hum can be an issue in every circuit and typically it depends alot more on the circuits surroundings than on the circuit itself.
jauu
Calvin
Finally got the thing packaged and patched into the system about 15 minutes ago, with a small SE tube amp doing tweeter duty, and a solid state on low-mids. Crossover is set at 5kHz.It will take me a while to get used to things, but the sound has tightened up.More reports will happen after some messing around, and when I'm sure I have the high-low balance straight.
re post14 & 15
The two schematics have the FETs running at different Id.
The p14 is running ~9mA and has a gate source voltage of -0.46Vgs.
I'd guess the Idss of this device ~15mA to 25mA
The p15 is running ~4mA through the same Idss device.
I think this 4mA bias is too low. Note, that gate source voltage is now -0.73Vgs
Experiment with the 150r source resistor in the CCS. Values between 50r and 150r are worth looking at.
Then adopt these optimum bias current resistors in the final p20 schematic, r8,10,12.
With the high pass filter, the input and output caps can usefully be made much lower in value. This would allow low value high performance caps to be economically adopted. I'm thinking that maybe polystyrene/foil or teflon/foil or polypropylene/foil. No metallised for the tweeter. For a 5kHz HP you could use F-3dB ~300Hz to 500Hz for the DC blocking caps.
The two schematics have the FETs running at different Id.
The p14 is running ~9mA and has a gate source voltage of -0.46Vgs.
I'd guess the Idss of this device ~15mA to 25mA
The p15 is running ~4mA through the same Idss device.
I think this 4mA bias is too low. Note, that gate source voltage is now -0.73Vgs
Experiment with the 150r source resistor in the CCS. Values between 50r and 150r are worth looking at.
Then adopt these optimum bias current resistors in the final p20 schematic, r8,10,12.
With the high pass filter, the input and output caps can usefully be made much lower in value. This would allow low value high performance caps to be economically adopted. I'm thinking that maybe polystyrene/foil or teflon/foil or polypropylene/foil. No metallised for the tweeter. For a 5kHz HP you could use F-3dB ~300Hz to 500Hz for the DC blocking caps.
Last edited:
I noticed the difference in ID when I posted the simulations and was too lazy to try and correct it at the time. I also thought that the sims were good enough to illustrate the point that the current source loaded source follower is lots better than the simple resistively loaded type.
I just ran a simulation for the current-sorce loaded version with higher ID, and there is about a 30% reduction in THD with the higher ID. The simulation model for the PN4393 runs out of gas at around 9ma.
The 150 ohm values used to size the source current in my first schematic (examples are R28, R24, R35, and R38) are really place holders. This is a point that I should have made clear from the start.
Since we are dealing with jfets, it's necessary to cherry pick parts for a given VGS vs. ID, or tailor the source resistor value for each jfet. I built a tester that allows me to look at VGS for ID of 1, 2, 5, and 10ma, as well as a setting for IDSS. In this case, I set the tester for 5ma, and started testing fets, looking for 8 parts that roughly matched in VDS for that drain current. I then used that VGS value to set the value of source resistance.
The PN4393 has a min IDSS of 5ma, so it might not be possible to get a close cluster of 8 parts with 10ma ID in a given lot. It'll depend on the batch. I have several sets of parts from various vendors and lots, and some batches are hotter than others. It would be possible to use a higher IDSS ranked part like the PN4392 to guarantee getting a good set of roughly matched parts at the higher drain current. This comes with the understanding that you will be running at a somewhat higher VGS, eating into your max signal swing a bit.
Another solution would be to dump the jfet current source and use something like a ring-of two source instead of the simple jfet current source. This approach is more complex, but conserves jfets (they're getting a bit thin on the ground these days), and eliminates all the matching fol-de-rol. I don't know that there is any advantage to matching the top jfets for VDS, as long as they are in the linear region for the ID that is selected. This can be determined with a simple IDSS test. I used parts left over from my matching session that still had significant VDS at 5ma.
The parts used are TO-92, and so have a junction-to ambient thermal resistance of about 200C/watt. This limits the amount of drain current you can run for a given VDS. Assuming a modest operating environment of 25C, this means that you can dissipate 125mW for a max junction temp of 50C. I wouldn't go past a junction temperature of about 70C. Assuming an ambient of 30C, 10ma drain current gets you to 60C. At 40C (Singapore with no air conditioning?), you're right at 70C. If you want to try a higher ID, you can reduce the VDS by lowering the supply voltage, with the understanding that you are reducing your maximum signal swing.
Executive summary - you are safe as houses at 5ma drain current and 15V VDS. At 10 ma, you're still ok, but getting warm with a high ambient temperature. For higher ID, you may want to consider reducing the supply voltage to limit dissipation.
I also considered going with smaller coupling capacitors for the high pass stage, but left the original values in out of laziness. I'm currently using some Roederstein metallized polypropylene caps to set the turnover frequencies. They are very tiny and extremely cheap (from a surplus source). The 1uF input coupling caps are metallized polycarbonate (surplus again), and the 3uF output caps are vintage high-current TRW (now ASC) metallized polypropylene (the TRW-35 series). If I were to do an premium version of this circuit, I'd consider using the mil-surplus Russian polystyrene caps available through Ebay and other sources. They are much larger that the caps I'm currently using, but might sound a little better. They are relatively huge, but there are several values of Russian teflon caps available that one could use in setting the turnover frequencies for both the high pass and low pass filters. 1-2 nf seems to be a comfortable working range, yielding comfortable values for frequency setting resistors.
I just ran a simulation for the current-sorce loaded version with higher ID, and there is about a 30% reduction in THD with the higher ID. The simulation model for the PN4393 runs out of gas at around 9ma.
The 150 ohm values used to size the source current in my first schematic (examples are R28, R24, R35, and R38) are really place holders. This is a point that I should have made clear from the start.
Since we are dealing with jfets, it's necessary to cherry pick parts for a given VGS vs. ID, or tailor the source resistor value for each jfet. I built a tester that allows me to look at VGS for ID of 1, 2, 5, and 10ma, as well as a setting for IDSS. In this case, I set the tester for 5ma, and started testing fets, looking for 8 parts that roughly matched in VDS for that drain current. I then used that VGS value to set the value of source resistance.
The PN4393 has a min IDSS of 5ma, so it might not be possible to get a close cluster of 8 parts with 10ma ID in a given lot. It'll depend on the batch. I have several sets of parts from various vendors and lots, and some batches are hotter than others. It would be possible to use a higher IDSS ranked part like the PN4392 to guarantee getting a good set of roughly matched parts at the higher drain current. This comes with the understanding that you will be running at a somewhat higher VGS, eating into your max signal swing a bit.
Another solution would be to dump the jfet current source and use something like a ring-of two source instead of the simple jfet current source. This approach is more complex, but conserves jfets (they're getting a bit thin on the ground these days), and eliminates all the matching fol-de-rol. I don't know that there is any advantage to matching the top jfets for VDS, as long as they are in the linear region for the ID that is selected. This can be determined with a simple IDSS test. I used parts left over from my matching session that still had significant VDS at 5ma.
The parts used are TO-92, and so have a junction-to ambient thermal resistance of about 200C/watt. This limits the amount of drain current you can run for a given VDS. Assuming a modest operating environment of 25C, this means that you can dissipate 125mW for a max junction temp of 50C. I wouldn't go past a junction temperature of about 70C. Assuming an ambient of 30C, 10ma drain current gets you to 60C. At 40C (Singapore with no air conditioning?), you're right at 70C. If you want to try a higher ID, you can reduce the VDS by lowering the supply voltage, with the understanding that you are reducing your maximum signal swing.
Executive summary - you are safe as houses at 5ma drain current and 15V VDS. At 10 ma, you're still ok, but getting warm with a high ambient temperature. For higher ID, you may want to consider reducing the supply voltage to limit dissipation.
I also considered going with smaller coupling capacitors for the high pass stage, but left the original values in out of laziness. I'm currently using some Roederstein metallized polypropylene caps to set the turnover frequencies. They are very tiny and extremely cheap (from a surplus source). The 1uF input coupling caps are metallized polycarbonate (surplus again), and the 3uF output caps are vintage high-current TRW (now ASC) metallized polypropylene (the TRW-35 series). If I were to do an premium version of this circuit, I'd consider using the mil-surplus Russian polystyrene caps available through Ebay and other sources. They are much larger that the caps I'm currently using, but might sound a little better. They are relatively huge, but there are several values of Russian teflon caps available that one could use in setting the turnover frequencies for both the high pass and low pass filters. 1-2 nf seems to be a comfortable working range, yielding comfortable values for frequency setting resistors.
To reduce confusion - in paragraph 6 of my last post the VDS should read VGS, where I'm talking about the necessity (or not) of matching the tops fets of a current-source loaded follower for VGS. Is should be sufficient to run the top fet far enough away from IDSS such that the gate is not driven positive for a reasonable signal excursion. This will depend on the load impedancce being driven. I chose 20k in my simulations as a compromise. I usually use a 47k inputt resistors for both my solid state and vacuum state amps.
For those interested, I'm using parts from the Roederstein KP1830 series for all my frequency determining caps. I happened on a local surplus source at a scandalously low price for 2% parts, so I couldn't resist. They are really small as well, with 5mm lead to lead spacing, so the board is fairly compact. The green Russian polystyrene caps available from the usual Ebay sellers are probably better. I may use them on my next pass for this design, as well as "FETWhite" followers.
Attached is a menengerie of buffer circuits to play with on sims (got LT Spice?) and on the bench. Circuits A-D are various single supply buffers, while E and F are configured to run on bipolar supplies. Some features can be mixed and matched between the various circuits for even more fun and games. Any one of these would be suitable for the unity gain filter circuits described here. There are more beasts that could possibly be added to the menengerie, but these are ones that have been mentioned here in passing, except for circuit D.
Attachments
Regarding the menengerie - Circuit "A" is a simple single ended source follower with a jfet current source load, biased for single supply operation, and the one I'm using in the current incarnation of my filter.
"B" is the same thing, but with a ring of two current source, so you can see how it's done.
"C" is the jfet version of the White cathode follower. The value for R15, which determines the drive to the bottom fet J5, is 1/gm, where gm is the transconductance of the bottom fet. For the PN4393, this is around 6-10 umhos, so R15 will be in the region of 100-150 ohms. A fet with higher transconductance like the 2SK170 will require a correspondingly smaller value for R15.
Circuit "D" is a current boosted source follower using an integrated darlington transistor (Q3). The ring of two current source (Q4-5) sets the total current going through J8 and Q3, and R23 apportions the current between J8 and Q3. The ring of two current source is used because jfets capable of linear operation at 20ma are pretty thinn on the ground, and the power dissipation would be too large for a TO-92 device anyway. Transistor Q4 should really be something a bit heftier than the 2N4401 shown in the schematic - a TO-92L package device like the MPSW06 would be suitable, and there are many excellent Japanese driver transistors in this package that would also work well.
Circuit "E" is circuit "A", tricked out for a bipolar supply.
Circuit "F" is a simple complementary buffer. R34 and R35 will probably need to be closer to 47 ohms for "BL" ranked devices.
Circuits A-E, if properly done, clock in at about 0.001% THD, overwhelmingly 2nd harmonic. Circuit F is about 10 times better in simulation, and is pretty well balanced, with about 30-40mV DC offset at the output. I don't know how this will translate to the real world, as the real devices will probably not be as well matched as the PSpice simulation models. Trying something similar in simulation with J176 and PN4393 or (J113 and J176) yields about 0.002% distortion and a lot more offset.
"B" is the same thing, but with a ring of two current source, so you can see how it's done.
"C" is the jfet version of the White cathode follower. The value for R15, which determines the drive to the bottom fet J5, is 1/gm, where gm is the transconductance of the bottom fet. For the PN4393, this is around 6-10 umhos, so R15 will be in the region of 100-150 ohms. A fet with higher transconductance like the 2SK170 will require a correspondingly smaller value for R15.
Circuit "D" is a current boosted source follower using an integrated darlington transistor (Q3). The ring of two current source (Q4-5) sets the total current going through J8 and Q3, and R23 apportions the current between J8 and Q3. The ring of two current source is used because jfets capable of linear operation at 20ma are pretty thinn on the ground, and the power dissipation would be too large for a TO-92 device anyway. Transistor Q4 should really be something a bit heftier than the 2N4401 shown in the schematic - a TO-92L package device like the MPSW06 would be suitable, and there are many excellent Japanese driver transistors in this package that would also work well.
Circuit "E" is circuit "A", tricked out for a bipolar supply.
Circuit "F" is a simple complementary buffer. R34 and R35 will probably need to be closer to 47 ohms for "BL" ranked devices.
Circuits A-E, if properly done, clock in at about 0.001% THD, overwhelmingly 2nd harmonic. Circuit F is about 10 times better in simulation, and is pretty well balanced, with about 30-40mV DC offset at the output. I don't know how this will translate to the real world, as the real devices will probably not be as well matched as the PSpice simulation models. Trying something similar in simulation with J176 and PN4393 or (J113 and J176) yields about 0.002% distortion and a lot more offset.
Thanks for sharing this! I'll be doing something similar once I finish getting my 3 way speaker put together. I'm doing the active crossover for the woofers to mid/tweeters, and passive for the mid to tweeters.
I admire your persistence in getting all this information posted. I'm sure it will come in very handy, for myself and others.
Have you done more listening to this yet?
John
I admire your persistence in getting all this information posted. I'm sure it will come in very handy, for myself and others.
Have you done more listening to this yet?
John
The schematic you draw is not a shunt regulator, it's a piece of wire (or maybe a pcb trace) directly from the 40V supply to the "30V" output, with a complex bunch of parts added for no usable or logical reason.
Now if you were to add a CCS in the 40V feed (or even a simple resistor), then it would suddenly transform itself into a working shunt reg
) and maybe work very well...Regards, Allen
Allen, please look again - R119 is the resistor you mean, working against Q17. Q5 and Q6 are a diff amp, fed by current source Q11. Current source Q12 feeds the LED voltage references D19 and D20. I ginned this thing up a couple of years ago, and it works ok. The board in question with filter and on-board supply has been sitting in my basement for a long time waiting for me to get to it, as I was diverted by the siren song of tubes. One thing that would be interesting to discuss regarding this circuit would be the relative merits of running the regulator circuit off its own output rather than off the raw input.
Scattered around in various places on Diyaudio (and working away in my living room) are the descendents of this regulator, which employ current source loading in place of the resistor and work a lot better. I've been doing a lot of simulation work recently on shunt regulators for use in active cathode bias. This came about as a response to SY's "Red Light District", which uses an LED array for bias in its output stage. I learned some interesting things, some of which will leak over into power supply applications. One of these days I should gather all the bits and pieces together for a unified thread on shunt regulator circuits, but it's a lot of work, and I have many other fish to fry. Some are sizzling in pans as I speak....
Scattered around in various places on Diyaudio (and working away in my living room) are the descendents of this regulator, which employ current source loading in place of the resistor and work a lot better. I've been doing a lot of simulation work recently on shunt regulators for use in active cathode bias. This came about as a response to SY's "Red Light District", which uses an LED array for bias in its output stage. I learned some interesting things, some of which will leak over into power supply applications. One of these days I should gather all the bits and pieces together for a unified thread on shunt regulator circuits, but it's a lot of work, and I have many other fish to fry. Some are sizzling in pans as I speak....
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.