Hi, I am looking for comments on a circuit and PCB Layout for a DAC gain stage. The idea is that you would use a dac with a passive I/V into this which would then provide the amplification. I am currently working on the DAC portion as well, but one could also use a HagDAC (http://www.hagtech.com/hagdac.html) or any number of other projects.
The circuit is a 5842 as a grounded cathode (I think that's the right term.) The plate load is a simple solid state CCS. This should provide a gain of about 43 and an output impedence of about 1.7K which is low enough for a preamp with a 50K pot.
The output coupling cap is connected to pad5 and pad6 and is mounted off the side of the board. The input coupling cap is likely unnecessary with some DACs in which case it could simply be jumpered. Otherwise there is room for a high quality cap there. (I may need an additional resistor to ground before the cap, but I am not really sure?) The heaters must be DC I would think, and this is the part I would especially like comments about as I am not sure the best way to lay them out, or if what I have done is fine.
As a side note, if I have boards made, I think that I will need to do at least 5. If this is the case, I'll likely have 3 or 4 extra. If anyone would want one, send me an email. Price would be whatever they cost to have made, which seems to be in the $15 range.
Oh, the 5842 seems to have 4 grid connections. I used which ever one seemed to be the closest to the input in the layout which is different for each side.
Also, C3 and C4 are not really polarized. Also also you the PS is seperate.
-d
The circuit is a 5842 as a grounded cathode (I think that's the right term.) The plate load is a simple solid state CCS. This should provide a gain of about 43 and an output impedence of about 1.7K which is low enough for a preamp with a 50K pot.
The output coupling cap is connected to pad5 and pad6 and is mounted off the side of the board. The input coupling cap is likely unnecessary with some DACs in which case it could simply be jumpered. Otherwise there is room for a high quality cap there. (I may need an additional resistor to ground before the cap, but I am not really sure?) The heaters must be DC I would think, and this is the part I would especially like comments about as I am not sure the best way to lay them out, or if what I have done is fine.
As a side note, if I have boards made, I think that I will need to do at least 5. If this is the case, I'll likely have 3 or 4 extra. If anyone would want one, send me an email. Price would be whatever they cost to have made, which seems to be in the $15 range.
Oh, the 5842 seems to have 4 grid connections. I used which ever one seemed to be the closest to the input in the layout which is different for each side.
Also, C3 and C4 are not really polarized. Also also you the PS is seperate.
-d
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
Where's the bypass for B+? Seems a little dangerous to count on the current sink to negate it's benefit. Why is C3 so physically large? You probably don't need any more than 0.1uF. Consider removing C1. The lack of it and slight cathode degeneration improve linearity and dynamics. Gain drops a tad, though, and output impedance rises. Boards look fine. Maybe you could cut it in half and use one for each channel? Unless you want the mirror effect.
And what is with this fake cascode current source? I keep seeing it used by everyone here. What's the benefit of such a connection? I don't get it. It's not a true cascode, it does NOT reduce output capacitance. Is it to compensate for Early Voltage? Or something to do with thermal transients from dynamic power loading on output transistor? Can someone please explain it?
jh
And what is with this fake cascode current source? I keep seeing it used by everyone here. What's the benefit of such a connection? I don't get it. It's not a true cascode, it does NOT reduce output capacitance. Is it to compensate for Early Voltage? Or something to do with thermal transients from dynamic power loading on output transistor? Can someone please explain it?
jh
Using a cascode as a constant current source allows you to sink a healthy anode current but have a high impedance load (roughly hfe1 x hfe2 x Re) with the result that distortion produced by the valve is very low indeed. And 5842 is a very linear valve in the first place. So once you drive a bit of constant current through it, you get a useably low output resistance with low distortion.
However, VHF oscillation is a distinct possibility. The HT needs to be decoupled with a 10n-100n capacitor with short leads/traces to the bottom of the cathode resistor and a surface mount grid-stopper resistor would be best. Most people would rely on the preceding device to decouple any input DC, but you definitely want an output decoupling capacitor.
However, VHF oscillation is a distinct possibility. The HT needs to be decoupled with a 10n-100n capacitor with short leads/traces to the bottom of the cathode resistor and a surface mount grid-stopper resistor would be best. Most people would rely on the preceding device to decouple any input DC, but you definitely want an output decoupling capacitor.
hagtech said:
In my head I was planning to put the final cap from a CLCRC filter near the board, and it was too large to fit on board. I do think I'll add a spot for the smaller decoupling cap suggested below, though.
hagtech said:
Space for a Jupiter 0.1uF cap.
hagtech said:
I think it depends on where I get it made. There are a few deals to get 5 "small" boards for a fixed rate which makes it in my interest to cram as much on each as possible.
hagtech said:
In my case, it is shamelessly stolen from a certain hi-fi kit manfacturer. I don't want to say who as people should buy their kit and not get the design from me.
EC8010 said:
Is it better to be closer to the cathode, or the top of the CCS, or right in between?
Also, surface mount over carbon? What sort of watt rating does the resistor need, or does it matter, and is it worth adding a spot for a plate stopper as well?
EC8010 said:
As the designer of the preceding device, I happen to know that it lacks a coupling cap. However, it seems to me that the raw dac ouput can't possibly have more than a fraction of a V of offset. Maybe it is possible to leave this out all together? I'll have to experiment.
SY said:
I think I'll leave the spot at any rate and maybe experiment both ways.
Thanks for the suggestions. I appreciate it.
-d
dsavitsk said:
Ideally, you want one end of the decoupling capacitor's short leads to go directly to the top of the current programming resistor in the cascode and the other directly to the bottom of the cathode resistor. If you can't do that, make the traces wide (reduces inductance).
Yes, SM over carbon. Reason being, on a PCB, you can butt the end of the resistor right up to the valve pin with no intervening track. It won't dissipate any power, but you'll probably prefer a larger resistor simply because it's easier to work with.
Okay, lots of work.
I made most of the changes suggested except the smd grid stopper. The problem is that there are traces running under the resistor, and using smd will make that portion of the layout a lot harder. I would rather move the resistors closer and mount them under the board if it is really necessary. But, if this is really emphatic, as in it is really necessary, then I will figure out a way to do it.
But, I added caps to decouple the PS. I also added smaller input caps in addition to the big ones.
Here is the changed board
However, I also made a version using a single 5687. The output impedence is similar, but the gain is a more tame 18 rather than 43.
Interestingly, this would also make a good preamp module, so I think it is more likely that I'll go ahead with this board and save the 5482 for later.
Here is thr 5687 board and partial schematic:
I made most of the changes suggested except the smd grid stopper. The problem is that there are traces running under the resistor, and using smd will make that portion of the layout a lot harder. I would rather move the resistors closer and mount them under the board if it is really necessary. But, if this is really emphatic, as in it is really necessary, then I will figure out a way to do it.
But, I added caps to decouple the PS. I also added smaller input caps in addition to the big ones.
Here is the changed board
An externally hosted image should be here but it was not working when we last tested it.
However, I also made a version using a single 5687. The output impedence is similar, but the gain is a more tame 18 rather than 43.
Interestingly, this would also make a good preamp module, so I think it is more likely that I'll go ahead with this board and save the 5482 for later.
Here is thr 5687 board and partial schematic:
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
Why would a grid stopper resistor have to be SM type? What is wrong with the classic carbon composition 1/4W or 1/2W leaded type? They work well at high frequencies (bulk type element), can handle power surge, and easily make voltage and layout spacing ratings.
Certainly the reduced inductance is of questionable merit, as the pins and leads within the tube itself far outweigh that of the resistor.
jh
Certainly the reduced inductance is of questionable merit, as the pins and leads within the tube itself far outweigh that of the resistor.
jh
Bear in mind that 5842 is very definitely a VHF/UHF valve, and it will oscillate at 50-100MHz given half a chance. It's not just 3mm of PCB trace, it's also 3mm or so of lead going through the board and bending up to the end cap. It all adds up. Of course, you can always swamp the whole thing by fitting a 1k grid-stopper, but that rather negates the purpose of choosing a low-noise valve. My point is that using SM instead of a leaded carbon film loses 6mm of lead which might just be the difference between oscillation and stability. There are no hard and fast rules here, but if it doesn't cost to have a theoretical advantage then why deliberately throw it away?
EC8010, your points are well made. Another approach where the SMD option is not available is to use several leaded resistors in parallel. For example, Steve Bench shows four 200 ohm resistors in parallel at each 417 grid. Lead inductance would then be 1/4 of a single equally-sized 50 ohm resistor. Leads should still be very short of course.
Another option, borrowed from the RF world, but maybe not so practical: If flat foil ribbon material can be found, say 0.25 inch wide, it can provide a low inductance lead from grid to resistor (as well as elsewhere). The foil would be soldered directly onto the tube socket grid pin and directly up to the body of whatever resistor you're using. I think I've seen ribbon-leaded RF resistors before, but I can't point to them right now. I see that Allen Wright uses silver foil ribbons for some of his wiring. I would think that cheaper copper foil (maybe from copper tape on a roll) would work well too.
In my experience with high-gm tubes, I have always just used a cheap leaded carbon-comp resistor wired close to the grid, and have been able to cure all oscillations with that as well as good PS bypassing and good grounding. I go for the smallest value stopper I can (and often that is none at all). Being a former RF engineer, I tend to wire things pretty tightly anyway, just out of habit. It helps to have a fast oscilloscope to see oscillations. I agree with EC8010 that slapping too large a value of grid stopper in place is an unneeded noise generator and a bandwidth limiter.
Another option, borrowed from the RF world, but maybe not so practical: If flat foil ribbon material can be found, say 0.25 inch wide, it can provide a low inductance lead from grid to resistor (as well as elsewhere). The foil would be soldered directly onto the tube socket grid pin and directly up to the body of whatever resistor you're using. I think I've seen ribbon-leaded RF resistors before, but I can't point to them right now. I see that Allen Wright uses silver foil ribbons for some of his wiring. I would think that cheaper copper foil (maybe from copper tape on a roll) would work well too.
In my experience with high-gm tubes, I have always just used a cheap leaded carbon-comp resistor wired close to the grid, and have been able to cure all oscillations with that as well as good PS bypassing and good grounding. I go for the smallest value stopper I can (and often that is none at all). Being a former RF engineer, I tend to wire things pretty tightly anyway, just out of habit. It helps to have a fast oscilloscope to see oscillations. I agree with EC8010 that slapping too large a value of grid stopper in place is an unneeded noise generator and a bandwidth limiter.
Banned
Joined 2002
dsavitsk said:
As Brian Beck says, it's all about good RF layout. I would suggest that spending a few hours of your time reading the HF and VHF project sections of either the ARRL or RSGB handbook would be an excellent investment of your time. I would feel pretty confident that I could lay out the PCB for 5842 and connect it to the power supply etc in such a way that it would not oscillate, so I'd expect to use my 350MHz oscilloscope simply to confirm that all was well.
Edit: I should say that I've made some fairly ghastly 5842 lash-ups that worked perfectly well with a single grid-stopper within 1mm of a grid pin. The multiple grid-stopper idea is very helpful in reducing inductance but not quite so feasible on a PCB.
Banned
Joined 2002
Two of my power amps and my headphone amp use 417A or the 5842 variant. I never had a problem with HF noise or oscillation, and I only use one pin. My 45 followed steve bench, i think, with 4 220R metal films. I removed them and didn't have an issue still with noise or oscillation.
Maybe I'm just a damn lucky bastard
Maybe I'm just a damn lucky bastard
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.