Holding an umbrella, unless I can find a Marine.
Actually, for reasons enumerated a few hundred (?) posts back, or was it in Sound Quality vs. Measurements, I like tubes for a variety of reasons. But low noise at low frequencies is not one of them.
Distortion cancellation with 10% matched n-ch is an illusion.
A few hours. I have more time than money to buy and match to 1 % the LSK devices. But I wouldn't bother doing that for a simple differential stage.
I would also appreciate if you could support your preference for duals. Of course, from your Guru position, you may chose to leave it there.
Good night.
To test high frequency response of a phono cartridge there are a couple of possible techniques that aren't too unreacable. First is spinning a test record faster. A 33.33 rpm test record that goes to 20 KHz will go to 46K at 78 RPM.
A technique developed by Ortofon is to use an accelerometer as an exciter. A suitable accelerometer can be had from eBay for not too much and an amp is pretty straightforward. You could chop a chunk out of an old record to get the vinyl compliance, It may be possible to get results to 100 KHz.
A technique developed by Ortofon is to use an accelerometer as an exciter. A suitable accelerometer can be had from eBay for not too much and an amp is pretty straightforward. You could chop a chunk out of an old record to get the vinyl compliance, It may be possible to get results to 100 KHz.
A 15 page article worth reading (Topology overview, theory, implementation, testing, references):
“A RIAA Equalized Preamplifier”
by Dimitri Danyuk and George Pilko, Kiev, Ukraine, 1988-1990
Dimitri really doesn’t enjoy stage lights.
Scott
I am trying to make my old hard disks and memory sticks work, so I can milk out old files. Some will never reveal their content (belly-up).
One retrieved folder so far confirms some of my memories. It may not be enough though to save my credibility
It contains many .xls files with FR measurements of a driven Shure M97 on air and in a stationary groove, varying electrical load. It will take a bit of time to put them in good form.
George
“A RIAA Equalized Preamplifier”
by Dimitri Danyuk and George Pilko, Kiev, Ukraine, 1988-1990
Dimitri really doesn’t enjoy stage lights.
Scott
I am trying to make my old hard disks and memory sticks work, so I can milk out old files. Some will never reveal their content (belly-up).
One retrieved folder so far confirms some of my memories. It may not be enough though to save my credibility
It contains many .xls files with FR measurements of a driven Shure M97 on air and in a stationary groove, varying electrical load. It will take a bit of time to put them in good form.
George
Here's some data to support my "distortion cancellation with 10% matched n-ch is an illusion" statement.
I took Mr. Cordell's jfet input stage, exactly as depicted in his VinylTrak article, used the NXP BF862 model (parametrized by Idss) and simulated the 2nd harmonic distortions (2HD) as a function of the Idss mismatch. The schematic and the simulation results are attached. In all cases, the 3rd harmonic is much lover than the second, and all other higher harmonics are virtually zero. So the 2HD pretty much defines the THD.
- There is very little relative 2HD frequency dependency in the audio band. I tried 1KHz and 20KHz, and the results are virtually identical.
- 2HD increase very quickly with the Idss mismatch.
- Ideal matching absolute 2HD were 0.25% for 50mV input at both 1KHz and 20KHz and 0.0047% and 0.0066% for 5mV input at 1KHz and 20KHz respectively.
- Taking the 0.9 Idss mismatch as a reference (it's the guaranteed Idss mismatch in the LSK389/489 data sheet) the 50mV 2HD increases by a factor of 2.8, while the 5mV 2HD increases by a factor of 9-10.
Conclusions:
- 2HD cancellation requires very close Idss matching. A 0.9 Idss ratio (as guaranteed by the much sought after monolithic dual jfets) is not good enough.
- Either a minimum 2HD is a critical design requirements (and then much closer matching, perhaps to 1%, is required), or the 2HD is considered low enough, anyway below the audible threshold, and then paying a premium for monolithic duals is useless.
Now, given these results, somebody please explain to a poor EE student why should one pay for the expensive dual monolithic. Modern devices from the same tube (or, even better, dual chip devices, which are always picked from neighbour locations on the wafer) are not far from the dual monolithic devices matching. Also the absolute distortion levels @5mV are already lower than the vinyl plus cartridge distortions, with or without device matching.
Somebody asked about mass production device matching. Set aside this is a DIY forum, if someone needs 10,000,000 matched pairs, I am sure OnSemi or NXP could provide sorting in tighter Idss classes, for a small premium fee per device. 10% matching as for the LSK389/489 would be a very easy task.
BTW, I had to manually match BF862's in 1% Idss for a university project that required very low input offset.
I took Mr. Cordell's jfet input stage, exactly as depicted in his VinylTrak article, used the NXP BF862 model (parametrized by Idss) and simulated the 2nd harmonic distortions (2HD) as a function of the Idss mismatch. The schematic and the simulation results are attached. In all cases, the 3rd harmonic is much lover than the second, and all other higher harmonics are virtually zero. So the 2HD pretty much defines the THD.
- There is very little relative 2HD frequency dependency in the audio band. I tried 1KHz and 20KHz, and the results are virtually identical.
- 2HD increase very quickly with the Idss mismatch.
- Ideal matching absolute 2HD were 0.25% for 50mV input at both 1KHz and 20KHz and 0.0047% and 0.0066% for 5mV input at 1KHz and 20KHz respectively.
- Taking the 0.9 Idss mismatch as a reference (it's the guaranteed Idss mismatch in the LSK389/489 data sheet) the 50mV 2HD increases by a factor of 2.8, while the 5mV 2HD increases by a factor of 9-10.
Conclusions:
- 2HD cancellation requires very close Idss matching. A 0.9 Idss ratio (as guaranteed by the much sought after monolithic dual jfets) is not good enough.
- Either a minimum 2HD is a critical design requirements (and then much closer matching, perhaps to 1%, is required), or the 2HD is considered low enough, anyway below the audible threshold, and then paying a premium for monolithic duals is useless.
Now, given these results, somebody please explain to a poor EE student why should one pay for the expensive dual monolithic. Modern devices from the same tube (or, even better, dual chip devices, which are always picked from neighbour locations on the wafer) are not far from the dual monolithic devices matching. Also the absolute distortion levels @5mV are already lower than the vinyl plus cartridge distortions, with or without device matching.
Somebody asked about mass production device matching. Set aside this is a DIY forum, if someone needs 10,000,000 matched pairs, I am sure OnSemi or NXP could provide sorting in tighter Idss classes, for a small premium fee per device. 10% matching as for the LSK389/489 would be a very easy task.
BTW, I had to manually match BF862's in 1% Idss for a university project that required very low input offset.
Attachments
Hi Waly,
I don’t know what I said to trigger your expression of resentment, but I assure you it was unintentional. I would hope that a civil discussion of two different views would not result in such hard feelings. I don’t consider myself a Guru – just a regular guy among a large group of very smart and experienced people on this forum.
Back to the technical matter, which is why we are all here.
Bear in mind that what matters in the symmetrical operation of a differential pair is the matching of the transconductances at the operating currents of the two devices. Those operating currents are usually maintained to be the same. Even with a 10% mismatch in gm, you will still get about 90% of the benefit of the cancellation of even order nonlinearity, which is certainly a lot more than an illusion.
But the good news doesn’t stop there.
Transconductance matching at a given operating current is not necessarily the same as saying that Idss of the two devices is the same. Indeed, transconductance at a given current in a JFET depends on Beta, the transconductance coefficient of the JFET, not Idss. We have, gm = 2 * SQRT(Beta * Id). For a given transconductance parameter Beta, gm is largely independent of Idss and Vt, and goes up as the square root of drain current. This is a good thing.
It would be nice if manufacturers specified matching of Beta, but they don’t. However, my experience with many JFETs is that Beta for a given device type and process varies rather little over Vt and Idss from device to device.
Thus, it is wrong to assume that a 10% difference in Idss will imply a 10% difference in gm at a given operating current. The latter mismatch will typically be much less.
I prefer duals because in practice they perform very symmetrically and I value my time. At the same time, others may choose to save a couple of dollars and match devices from the same tube, and they also will achieve excellent results. It’s just a personal choice.
By the way, we are not talking about phono preamps you purchase at Best Buy in this thread. If you are worried about spending a couple of dollars on a dual JFET (or an AD797, for that matter), you may be participating in the wrong thread.
Cheers,
Bob
Bob,
This is J.C.'s space, feel honored to be mistreated. It means that others feel you are at the highest levels here.
Waly seems to be new and did duck his head down after one fire fight, so he does have some common sense.
Now my usual disagreements with you come from using jargon, that we assume means the same to the other person as it does to us.
I try (and regularly fail) to define a term when using the symbol for it, when doing tutorials. For example what do you mean by "gm" and "transconductance?"
Of course you still get PIM (Passive Intermodulation Distortion) wrong,as you seem to think it has something to do with phase versus level!
ES
I did not say that. I simulated a 10% Idss mismatch, not a 10% Gm mismatch. Gm~SQRT(Id*Idss), the Gm ratio stands as the square root of Idss's (since the operating currents are the same), so a 10% Gm mismatch would have a 3 times larger impact than a 10% Idss mismatch.
In practice, the devices in a dual part may be much closer than 90%. I believe much of what you are maintaining from a practical perspective is related to this. However, I think you agree that relying on better than data sheet parameters is not good engineering practice.
I suppose I'm missing here something. Beta is an intrinsic parameter of the device. FWIW, the designer doesn't care about the grueling details like carrier mobility, channel length/width ratio, etc... and relies on simple equations like the one I quoted above, tying transconductance to the static operating point and the external electrical parameters. Another version is Gm=2*Idss/Vp - I think saying that Gm is independent on both Idss and Vp is a stretch. At least two quantities of the Beta, Idss, Vp triplet are independent and must be specified.
Again, I did not say that, see above. The simulation results are based on Idss mismatching, hence (being a rather theoretical approach) a three times lower Gm mismatch.
If I would be in your Guru position, I would be more careful in exposing personal preferences. They may be interpreted as sneering or handwaving (which I did, and apologize).
You are on the other side of the pond - you have no idea how much would cost to bring a few LSK's to the UK. Believe me, it's ridiculous. I was simply trying to bring to the lesser than NA DIYers that dual jfets are largely overrated and hence those shekels could be used in a more effective way. One of these ways is to get the cascoded or duals I quoted from OnSemi, available at Digikey for pennies. Do you really believe that the LSK's would make any significant difference in a MM preamp, compared to the OnSemi devices from the same tube? Not to mention the possibility to get a much more compact SMD build, hence less EMI and hum pickup. Think how much space would save using those SOT23 jfet + bipolar cascode...
BTW, to address here the jcx concern, there's not a big problem to perfectly match n-ch and p-ch devices (assuming a large enough population is allowed). It is both the Idss, Vp (and hence Gm) and the Ciss, Crss that can't be simultaneously matched. Same applies to power mosfets - matching for transconductance mismatches Ciss and Crss.
Last edited:
Waly, you make a good point, but your 'criticism' of matched pairs was a bit overdone. You have a 'special' problem, i.e. low money and distance from the supplier, so the extra effort that you make is worthwhile. Not so for many here in the USA and Canada.
However, both Parasound and Constellation, two companies that I design products for, have to hand match, these days, because we can't get the Toshiba k389 and j109 in quantity, anymore. We hand match K170 and J74's by the hundreds.
The REAL match is in the voltage offset between the two devices, differentially, which is even more difficult to do than Idss matching, which is easy, IF you have 100's of devices and plenty of time, but differential voltage offset is something that monolithic construction does very well.
How important is it? A little 2'nd harmonic may spoil the measurement results, but it is seldom, if at all, audible.
However, both Parasound and Constellation, two companies that I design products for, have to hand match, these days, because we can't get the Toshiba k389 and j109 in quantity, anymore. We hand match K170 and J74's by the hundreds.
The REAL match is in the voltage offset between the two devices, differentially, which is even more difficult to do than Idss matching, which is easy, IF you have 100's of devices and plenty of time, but differential voltage offset is something that monolithic construction does very well.
How important is it? A little 2'nd harmonic may spoil the measurement results, but it is seldom, if at all, audible.
If you read the datasheet this is their take on the AM radio socket filler. Does anyone have a reference design, I don't see how they have not fully integrated something so simple as an AM radio?
Thanks Demian
The accelerometer has to be characterized first.
Here is the B&K article (pdf pages 27-36)
http://www.bksv.com/doc/TechnicalReview1976-2.pdf
It answers many of the questions here (and generates a few more
)George
>Edit. This is another valuable article of them (test records and vinyl issues)
http://www.bksv.com/doc/TechnicalReview1962-3.pdf
Last edited:
- Status
- Not open for further replies.