In a thread about 5842 GG gain blocks, Brian Beck posted the following.
Brian has provided some important info. If a 4.99 Ohm I/V resistor is employed, a 19.96 mV. (peak) signal is produced. A 6922 with its sections in cascode can yield the gain needed to raise that feeble signal to the level usually associated with CDPs.
Recently, Eddie Vaughn was kind enough to provide me with a set of operating conditions for 6922 cascodes.
Is there any interest?
Brian Beck said:
Brian has provided some important info. If a 4.99 Ohm I/V resistor is employed, a 19.96 mV. (peak) signal is produced. A 6922 with its sections in cascode can yield the gain needed to raise that feeble signal to the level usually associated with CDPs.
Recently, Eddie Vaughn was kind enough to provide me with a set of operating conditions for 6922 cascodes.
Is there any interest?
Eli,
A good topic for a thread. I waited for others to chime in, and seeing nothing from overnight, I thought I'd toss out a few thoughts.
If you use a very small resistor as a pseudo- I/V converter at the output of a current-output DAC, then the following amplifier stage starts to share some similarities to a phono stage, without the RIAA equalization of course. The input levels are about the same as a moving-magnet phono stage, and the gain required will be similar (as at 1KHz in the RIAA stage). A typical passive EQ RIAA stage (MM) has a low frequency gain of about 60dB, a mid-frequency gain of about 40 dB, and a high frequency gain of about 20dB. Here we need something like 40 to 50dB across the board, depending on the I/V resistor selected and the DAC chip specifics, of course. So in a general way, we can borrow techniques, particularly as regards low-noise behavior, from existing RIAA stage designs. But we can detune the raw gain from 60dB to maybe 40 to 50dB. This suggests two triode stages, with or without feedback, plus perhaps a CF for low-output resistance.
I must say that we have two conflicting desires here. On the one hand, we want to make the I/V resistor very small to keep the DAC within spec, but on the other hand we want the signal voltage to be high enough to keep noise from being a problem. As we drop the resistor value, the gain must go up, and with it the noise. The noise from a typical RIAA stage will be much higher than that from a DAC. So this technique of a using a very small resistor for I/V sacrifices S/N ratio performance, but perhaps could sound pretty good otherwise. As you know, I'm no fan of FETs, but here we could consider a FET/triode cascode for lowest noise. Since the FET's gate is driven by a 5-ohm source resistance, the effect of the variable gate capacitances is knocked out.
These conflicting goals remind me of an analogy: It's like riding the brakes in a car traveling at a steady highway speed. As we reduce the resistor value to make the DAC happy, it's like pressing the brake pedal harder. To keep the same speed (output voltage), we must press the accelerator pedal harder (gain) to compensate. Fuel efficiency (low noise) suffers.
This is why I suggested considering the tube "opamp" with feedback as a way to allow the DAC to see a virtual ground, while preserving good S/N. So we have a trade-off between S/N ratio and the use of feedback. Which approach might sound better is the big question, and it all would depend on many factors.
A good topic for a thread. I waited for others to chime in, and seeing nothing from overnight, I thought I'd toss out a few thoughts.
If you use a very small resistor as a pseudo- I/V converter at the output of a current-output DAC, then the following amplifier stage starts to share some similarities to a phono stage, without the RIAA equalization of course. The input levels are about the same as a moving-magnet phono stage, and the gain required will be similar (as at 1KHz in the RIAA stage). A typical passive EQ RIAA stage (MM) has a low frequency gain of about 60dB, a mid-frequency gain of about 40 dB, and a high frequency gain of about 20dB. Here we need something like 40 to 50dB across the board, depending on the I/V resistor selected and the DAC chip specifics, of course. So in a general way, we can borrow techniques, particularly as regards low-noise behavior, from existing RIAA stage designs. But we can detune the raw gain from 60dB to maybe 40 to 50dB. This suggests two triode stages, with or without feedback, plus perhaps a CF for low-output resistance.
I must say that we have two conflicting desires here. On the one hand, we want to make the I/V resistor very small to keep the DAC within spec, but on the other hand we want the signal voltage to be high enough to keep noise from being a problem. As we drop the resistor value, the gain must go up, and with it the noise. The noise from a typical RIAA stage will be much higher than that from a DAC. So this technique of a using a very small resistor for I/V sacrifices S/N ratio performance, but perhaps could sound pretty good otherwise. As you know, I'm no fan of FETs, but here we could consider a FET/triode cascode for lowest noise. Since the FET's gate is driven by a 5-ohm source resistance, the effect of the variable gate capacitances is knocked out.
These conflicting goals remind me of an analogy: It's like riding the brakes in a car traveling at a steady highway speed. As we reduce the resistor value to make the DAC happy, it's like pressing the brake pedal harder. To keep the same speed (output voltage), we must press the accelerator pedal harder (gain) to compensate. Fuel efficiency (low noise) suffers.
This is why I suggested considering the tube "opamp" with feedback as a way to allow the DAC to see a virtual ground, while preserving good S/N. So we have a trade-off between S/N ratio and the use of feedback. Which approach might sound better is the big question, and it all would depend on many factors.
To reduce the digital garbage going into the tube stage, and as part of a gentle reconstruction filter, we can add a cap in parallel with the 5-ohm I/V resistor (using your example value). The cap has to be pretty large to make a given time constant with only 5 ohms. If we put a 1uF cap in parallel with the 5-ohm resistor, we'll get a pole (-3dB) at about 32kHz. This cap diverts digital noise current away from the 5-ohm resistor, reducing the resultant I/V voltage that would get into the tube stage and cause IMD. For a sharper filter, we can add a series L followed by a shunt C after the 5-ohm resistor. And we could take the passive filtering even further still. But I would suggest at least a single shunt cap to reduce the slew rate of noise going into the tube stage.
Brian,
I agree, the low value I/V resistor regime is similar to that of a MM phono cartridge.
I LIKE the low pass cap. at the I/P idea. A 1 muF. RTX MultiCap paralleled by a 100 pF. silvered mica part should do an excellent job of shorting HF crud to ground.
6922s are definitely quiet enough for MM phono section service. I suggested a 6922 cascode, as 100% of the gain needed can be obtained in a single block. A 10 KOhm load resistor in combination with a DC coupled voltage follower seems correct.
FWIW, I have some thoughts about the "hard" sound that's been associated with cascodes. I suspect parasitic oscillation is a good part of the problem. The upper triode in a cascode is, in fact, operating GG. That (IMO) means: no grid stopper resistor, a high quality cap. connection to ground, and RF suppression on the heater leads. The common cathode lower triode needs (sic) both grid and plate stopper resistors. A 1 KOhm Carbon comp. on the grid and a 100 Ohm Riken on the plate "feels" right to me.
Oh yeah, since noise is an issue, expensive Vishay S102s in the 4.99 Ohm I/V positions may be a rational selection.
I agree, the low value I/V resistor regime is similar to that of a MM phono cartridge.
I LIKE the low pass cap. at the I/P idea. A 1 muF. RTX MultiCap paralleled by a 100 pF. silvered mica part should do an excellent job of shorting HF crud to ground.
6922s are definitely quiet enough for MM phono section service. I suggested a 6922 cascode, as 100% of the gain needed can be obtained in a single block. A 10 KOhm load resistor in combination with a DC coupled voltage follower seems correct.
FWIW, I have some thoughts about the "hard" sound that's been associated with cascodes. I suspect parasitic oscillation is a good part of the problem. The upper triode in a cascode is, in fact, operating GG. That (IMO) means: no grid stopper resistor, a high quality cap. connection to ground, and RF suppression on the heater leads. The common cathode lower triode needs (sic) both grid and plate stopper resistors. A 1 KOhm Carbon comp. on the grid and a 100 Ohm Riken on the plate "feels" right to me.
Oh yeah, since noise is an issue, expensive Vishay S102s in the 4.99 Ohm I/V positions may be a rational selection.
Sheldon said:
Sheldon,
A custom trafo might work. In order to avoid caps., the DAC chip has to work directly into the trafo's primary. That means the trafo has to present a 5 Ohm load to the chip. The secondary is going to have specific resistive loading requirements. Can the balance between frequency response and ringing be maintained?
Eli Duttman said:
I suspect you're right about instability being a factor in the hard sound. Also, the voltage gain equation for the cascode is dependent on the more non-linear parameter gm rather than on the more constant parameter mu, as it is in the case of the common-cathode stage with a large load resistor. In this regard it's similar to a pentode and will make more higher-order distortion components that can sound hard if they're large enough. For our application, the signal levels coming off that 4.99 ohm resistor are pretty small, so a cascode has promise (as Allen Wright would no doubt agree).
I agree that the 4.99-ohm resistor ought to be of good quality, but more because we want a low voltage coefficient of resistance to keep distortion low. This 5-ohm noise resistance will be dwarfed by the first stage triode's noise resistance. A 6922 might have an equivalent noise resistance of 300 ohms if it's biased richly. That will make at least 18dB greater noise than a 5-ohm resistor, not to mention grid stopper noise contributions, 1/f noise, etc.
Eli Duttman said:
Oh yeah, these foil resistors are not only good enough, but I think many people would think that they're overkill. So as not to scare off others, can we agree that a good old 1% metal film will work here also? Either way the noise created by the first active device will far exceed any excess noise created by either kind of 4.99 ohm resistor.
A couple of little things to add:
Here's a nice circuit from Gordon Rankin that might be worth considering:
http://www.diyhifi.org/forums/files/simivtube_143.jpg
Also, Sowter makes transformers specifically for this application. http://www.sowter.co.uk/dacs.htm AudioNote does too, but I don't know if they are available to the hobbyist w/o buying a kit. Sowter does suggest putting the I/V resistor on the secondary which would mean the only DCR would be that of the primary.
How about a choke, or an autoformer instead?
Whether they should or not, the I/V resistors make a huge difference in sound Cheap metal films tend to sound bad. Tants and Rikens are nice, but my favorites thus far are ABs. Maybe it is just the structure of the distortion ...
Here's a nice circuit from Gordon Rankin that might be worth considering:
http://www.diyhifi.org/forums/files/simivtube_143.jpg
Also, Sowter makes transformers specifically for this application. http://www.sowter.co.uk/dacs.htm AudioNote does too, but I don't know if they are available to the hobbyist w/o buying a kit. Sowter does suggest putting the I/V resistor on the secondary which would mean the only DCR would be that of the primary.
How about a choke, or an autoformer instead?
Whether they should or not, the I/V resistors make a huge difference in sound Cheap metal films tend to sound bad. Tants and Rikens are nice, but my favorites thus far are ABs. Maybe it is just the structure of the distortion ...
dsavitsk said:
Huh? With all due respect to Gordon, either we have circuit that is out of the context of what we’re discussing here, or this circuit misses the boat. Just because the input goes into the cathode doesn't mean that it will see a particularly low input impedance such as we need for the DAC. The plate load impedance is “seen” through the cathode, divided by mu+1. In this circuit, if we load the output transformer with 100K, a typical value for a line stage, then that value is transformed through the output transformer to 2.5M at the plate. In parallel with that is the 225H choke, which at 1 KHz presents 1.4M of inductive reactance to the plate. The combined load will be a complex impedance of very roughly 1M. This value is divided by mu+1 and gives us roughly 30K at the cathode. That’s the load the DAC sees, and it’s also completely dependent on the load on the secondary of the output transformer, as well as being frequency dependent. Compare that to the 5 ohms needed. Sanity check: if the DAC can output +/-1mA (typical), then the voltage will swing by +/- 30 volts at the DAC output. “Pop” goes the DAC. If the DAC doesn’t go “pop” right away, the voltage swing on the plate would be in severe clipping. Gordon has to have been assuming something else altogether. Or, maybe this was just a paper exercise.
JoshK said:
Josh,
The I/P impedance of the cascode is in parallel with the 4.99 Ohm I/V resistor. Since that impedance is several orders of magnitude greater than the I/V resistor, it is effectively "invisible" to the DAC chip.
The O/P of the cascode presents its own issues. A DC coupled voltage follower is needed to buffer the high O/P impedance. Also, PSRR is non-existent in cascodes. So, the B+ must be WELL filtered and regulated.
Sheldon,
Your schematic is reasonably consistent with Allen Wright's phono section FET cascode.
Stopper resistors are needed on the FET's gate and the CF's grid. DC couple the CF's grid to the cascode and adjust its cathode resistor accordingly. Also, you left the low pass cap. across the I/V resistor out.
As Mr. Beck indicated, the gain of a cascode is tied to the transconductance of the lower element. The 10 KOhm load resistor value I gave was based on 6922 sections serving in both the top and bottom positions. A different value is needed with a FET in the bottom position.
The voltage divider that sets the cascode's idle current could be made a pot. A 100 Ohm Riken at the FET's drain would do double duty as a stopper and a test point.
I think this circuit has potential. AW's FET cascode is quiet enough for use with LOMC phono cartridges. The proverbial church mouse is noisier.
BTW, AW sells closely matched pairs of low noise JFETs.
"Just what the doctor ordered."
Your schematic is reasonably consistent with Allen Wright's phono section FET cascode.
Stopper resistors are needed on the FET's gate and the CF's grid. DC couple the CF's grid to the cascode and adjust its cathode resistor accordingly. Also, you left the low pass cap. across the I/V resistor out.
As Mr. Beck indicated, the gain of a cascode is tied to the transconductance of the lower element. The 10 KOhm load resistor value I gave was based on 6922 sections serving in both the top and bottom positions. A different value is needed with a FET in the bottom position.
The voltage divider that sets the cascode's idle current could be made a pot. A 100 Ohm Riken at the FET's drain would do double duty as a stopper and a test point.
I think this circuit has potential. AW's FET cascode is quiet enough for use with LOMC phono cartridges. The proverbial church mouse is noisier.
BTW, AW sells closely matched pairs of low noise JFETs.
"Just what the doctor ordered."
Sheldon said:
All of this is "brainstorming", at least for now.
Loose the grid stopper on the upper triode of the cascode. Remember, the upper triode in a cascode operates GG. The voltage divider at the buffer's I/P grid has to go too. Wire the ends of a pot. between B+ and ground and connect the wiper to the upper grid of the cascode. Put a 100 Ohm Riken between the cascode's lower plate and upper cathode, as a stopper/test point. LED bias for the cascode will work, but I prefer an unbypassed metal film resistor. Local current NFB from degeneration is (IMO) a good thing here. The cascode's load resistor gets tweaked to yield the amount of gain needed.
BTW, a 6922 for the cascode and an ECC99 for the buffer looks good to me.
Fixed the pot. The resistors on the buffer grid, Broskie adds as safety resistors, but I may have missapplied them. I left them out for clarity here. So more like this?
Edit: On further checking, realized that John has beat me to it (no surprise there). http://www.tubecad.com/2004/blog0013.htm
Edit: On further checking, realized that John has beat me to it (no surprise there). http://www.tubecad.com/2004/blog0013.htm
Attachments
Sheldon said:
Getting close! There is no need for a fixed resistor connecting the "top" of the cascode's current set pot to B+. The value of the current set pot. should be no less than 1 MOhm. The cap. to ground gets connected directly to the upper triode's grid. Think GG!
PSRR in the White CF buffer is good enough to allow connection to well filtered, but unregulated B+. Reserving the regulator to feeding only the cascodes decouples the cascodes from the buffers.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.