G9 and its frequency response ?? (impedance matching?)

[Kingston]

don't help if the input stage is killing the bass before it gets to the first Triode.

like I said, almost negligible effect in comparison to fixing the output.

http://www.groupdiy.com/preamplifiers/g9-and-its-frequency-response-%28impedance-matching%29/msg394779/#msg394779

I wasn't using RMAA back then, but the lower graph reflects your GIX2 configuration.

[edit]

some pages earlier there was another graph with the OEP as output (stock configuration). http://www.groupdiy.com/preamplifiers/g9-and-its-frequency-response-%28impedance-matching%29/msg393728/#msg393728

A lot flatter than lundahl, I didn't remember that. Notice how the bass has a gentle slope, but no abrupt roll-off like lundahl, showing how little the input transformer affects things here.

Ignore all the edcor hacking stuff, I later measured the phase response and it was completely wacked. But a non-hacked 2:1 edcor WSM or XSM would work ok.

 

[gemini86]

Yes the oep is wired both PRI and sec in series so changing the sec (which is actually the primary, since it's being used backward) to parallel will give 4:1.

 

[Kingston]

As a sidenote, those measurements and preferences were too many years a go, stuff I wrote about caps should be simply ignored. I've learned not to blindly trust my ears since.

And it could have been 2010 when I put the OEP output back in and it has stayed there since. 2:1 configuration with ECC88 driving it. 4:1 with 12AU7/ECC82 would be equally valid if you can live with the doubled insertion loss.

OEP has a nice (literal) ring to it - says the scope - but it just sounds better than LL5402 in this project.

 

[bruce0]

I did read up on the higher current tube and cap changes and increased drive etc.  But all of my testing (and many of the comments back in 2010, point to an input stage rolloff (many folks noted the better low end using the DI input ).  The charts show a big rolloff, and there is some but my test gear has a bad LF rolloff in it too, so the rolloff in these charts is repeatable and precise, but not accurate with respect to the real performance of the device under test.  But comparing the traces, the difference on the OEP A262a3E of putting some load on the secondary and lowering the mic impedance is clear.

The HF peak tamed by loading the input trafo (not the output load) and lowering the input impedance.  And halving R5 makes the pad 32dB instead of 26dB but that's ok with me a reasonable pad for Line to Mic level use.

In any case, here are the sweeps with the output left alone loaded by the built in 10K resistors in parallel, and the input terminated in various ways (strapped across the back of the A262A3E).  And as you can see that input trafo is very susceptible to differences in Mic impedance.

The light blue line is the configuration I settled on.

The second sweep is trying different impedances on the source (for instance a 600 Ohm lin source, and 3 different mic's) to see how well the A262a3e loading works, and it works acceptably as far as I am concerned. All traces on this sweep are with: 130K ohms between TR1 B4 and GND, and with a 680pF cap in series with 100K parallel to that.  I think that is where I will leave it for a while.

I will try to measure the low frequency response at some point, or at least subtract out the effect of my test interface low end rolloff.

Edit:
The LL Line level sweeps are done at +4dB input (sometimes more), 0dB output
The Mic Level sweeps are done at -24dB input, 0dB output

My sense is that the Output section doesn't really have a problem with any of the different transformers I have tried (1 hand wound, 2 lundahls, and an OEP ).  I do want to look into the other input transformers I am using, but all "measure" like they don't need the termination.

Some OEP datasheets for A262A3E use a 6K load, but since we want to present 1-3K to the mic, I didn't think that was a good idea.

 

[bruce0]

Kinston:

I saw that chart before... and never really understood it, perhaps I don't understand the scale (isn't it a dB scale?).  From what I could see you were down 5 dB at 200 hZ and rolling off steep even before i tried to "optimize" this build, i was only down 0.5db at 200hZ. 

Can you explain why you say there is negligible effect in improving the input stage?  I think you can see that properly terminating the OEP input picks up moves it from -4dB to -2dB at 50 Hz and halves the rolloff at 200hZ even with these cheap little OEP's.  Messing with later stages to fix (eq) doesn't seem right, better to capture the signal before amplification stages rather than recreate what it should have been.  But I am not an expert in all this.



bb

don't help if the input stage is killing the bass before it gets to the first Triode.

like I said, almost negligible effect in comparison to fixing the output.

http://www.groupdiy.com/preamplifiers/g9-and-its-frequency-response-%28impedance-matching%29/msg394779/#msg394779

I wasn't using RMAA back then, but the lower graph reflects your GIX2 configuration.

[edit]

some pages earlier there was another graph with the OEP as output (stock configuration). http://www.groupdiy.com/preamplifiers/g9-and-its-frequency-response-%28impedance-matching%29/msg393728/#msg393728

A lot flatter than lundahl, I didn't remember that. Notice how the bass has a gentle slope, but no abrupt roll-off like lundahl, showing how little the input transformer affects things here.

Ignore all the edcor hacking stuff, I later measured the phase response and it was completely wacked. But a non-hacked 2:1 edcor WSM or XSM would work ok.

 

[Kingston]

I saw that chart before... and never really understood it, perhaps I don't understand the scale (isn't it a dB scale?).  From what I could see you were down 5 dB at 200 hZ and rolling off steep even before i tried to "optimize" this build, i was only down 0.5db at 200hZ. 

Can you explain why you say there is negligible effect in improving the input stage?

first of all, I've never used the OEP input transformers so if there is a problem (seems to be), I wouldn't know.

The scale in those FFT plots isn't accurate down to dB. I wish I had used RMAA at the time, but there is a reference "perfectly flat" AD/DA loop to compare and the unfixed lundahl output shape certainly reflects your RMAA measurement quite perfectly.

And you can definitely see the effects of fixing the output stage and even the output cap low frequency bump. The reason I'm saying these input transformer tweaks are somewhat futile is that when you fix the output stage, the response is flat down to maybe 20hz (sans the LF cap bump). At that point any tweaks at the input stage are effecting something like 10-20hz.

Talking about the lundahl input only, can't comment on OEP.



re: G9 as a general purpose unit.

the unit has a specific two step low cut switch by default. Clearly it tries to be general purpose, giving options like this. Why would you not also want a flat frequency response. Handicapping the output stage on purpose is beyond my understanding.

 

[gemini86]

If I needed a perfectly flat response, I'd use a 9K.

I'm with bruce on this one, if the input aint right, let's fix it first, then see what the output stage needs to be doing.

 

[Kingston]

if the input aint right, let's fix it first

I can't seem to make myself clear and understood. The input stage is perfect with LL1538. There is nothing to fix.

This is what you can expect with a proper output stage. ECC88 to OEP 2:1, 10uF output cap. Lundahl output would be flat up to even 30khz and channels would match better, but hey let's leave a trace of sound to the unit shall we. Left channel is using the stock line input pad to LL1538. Right channel audio goes directly to DI input - in other words, no input transformer.

Audio source is a 600-ohm transformer DAC output stage. My own DAC output testing rig that I had to use the get the DI input matched to the line input and not clipping. The high frequency wavey line is a DAC output transformer ringing artifact because it doesn't like to work into the very high impedance DI input. But we can ignore that, this demonstration is for the bass response only and never in the real world we would take a line level output to any DI input anyway.

I prefer the dedicated low cut switch does all the low cutting work when required, and not some design related artificial handicap that you can't turn off.

 

[gemini86]

Sorry, I wasn't meaning to dismiss your point Kingston. I don't doubt there's a problem with the output stage, as designed. It just seemed to me that you don't believe there's improvements to be made on the input, when using the oep transformers. I think it's very much worth the effort to optimize the circuit for use with the oep transformers, when there are so many of us with shallow pockets.

 

[Kingston]

It just seemed to me that you don't believe there's improvements to be made on the input, when using the oep transformers.

In fact I find it impossible, with all the measurement data available. Seems to me all the impedance matching at the input is doing nothing, a dB here, another there. Basically within normal parts tolerance we can expect in any transformer I/O.

Looking at GIX1 - GIX3 shows Lundahl basically equal to OEP, besides high frequency ringing. And my own data confirms this, with input transformer eliminated from the equation completely.

Trying to fix the input stage at this point is kind of like going to a dentist when you have hemorrhoids.

 

[Kingston]

Since some people and definitely you two are confined to GIX51x constraints, a valid solution would be a 4:1 output transformer (both the OEP and Lundahl can do this). It only involves cutting traces, or making an adapter rig. Trade off is 6dB less gain. No big deal, plenty to spare.

 

[bruce0]

if the input aint right, let's fix it first

I can't seem to make myself clear and understood. The input stage is perfect with LL1538. There is nothing to fix.

This is what you can expect with a proper output stage. ECC88 to OEP 2:1, 10uF output cap. Lundahl output would be flat up to even 30khz and channels would match better, but hey let's leave a trace of sound to the unit shall we. Left channel is using the stock line input pad to LL1538. Right channel audio goes directly to DI input - in other words, no input transformer.

Audio source is a 600-ohm transformer DAC output stage. My own DAC output testing rig that I had to use the get the DI input matched to the line input and not clipping. The high frequency wavey line is a DAC output transformer ringing artifact because it doesn't like to work into the very high impedance DI input. But we can ignore that, this demonstration is for the bass response only and never in the real world we would take a line level output to any DI input anyway.

I prefer the dedicated low cut switch does all the low cutting work when required, and not some design related artificial handicap that you can't turn off.

Through the DI, I see.  I will make that test and see what I get.  (Funny... the DI use explains why you don't have a problem in the Input stage, you are a guitarist... and that could explain the obtuseness as well (forgive this little attempt at humour)>.  You are testing through the DI, if you make tests that don't involved the DI, you will involve the input transformer in the input stage.

I will try my sweeps through the DI, avoiding the Transformer, just for fun.  But it won't help me tune the input trafo.

 

[Kingston]

You are testing through the DI, if you make tests that don't involved the DI, you will involve the input transformer in the input stage.

*sigh*

Left channel is with the input transformer. I'm not so sure I'm the obtuse one here.

 

[bruce0]

Ooops... missed that.

Still, I think I will run mine through a DI.  This design has some untested aspects, and the results I am seeing don't align perfectly with what you were seeing prior to your output stage changes.

One of my concerns is to the current limitations on the B+ circuit.  The stock G9 has a big reservoir cap (100uF) feeding the driver/buffer stage tube plate directly.  Stock the GIX51X design has 20uF, and with my filter mods and an upgraded cap mine has 22uF (because i isolated the PSU reservoir with a filter, I upped the cap size).  I can add more cap there, but it is a bit messy (Axial Cap, on the surface of the board) due to size constraints, and I didn't think it was needed.

If I test with the DI and see the problem you had, well I guess that means the driver stage is fine except for known issues you pointed out related to SRPP stage design and tube choice.  But if I see something worse, well maybe I can get an improvement with the B+.

I do note that your low end response with a LL1538 input tranny is good, which is surprising, since mine (net of output stage issues) is still quite a bit worse than that (currently the OEP is better, oddly).  But I need to match your test conditions (600 ohm input impedance), which I am in the process of setting up.  You mention the 1538, original G9 schemo called for the 1528.  I am assuming you are using the 1538.

Sorry to make you sigh.

bb

You are testing through the DI, if you make tests that don't involved the DI, you will involve the input transformer in the input stage.

*sigh*

Left channel is with the input transformer. I'm not so sure I'm the obtuse one here.

 

[Kingston]

One of my concerns is to the current limitations on the B+ circuit.  The stock G9 has a big reservoir cap (100uF) feeding the driver/buffer stage tube plate directly.  Stock the GIX51X design has 20uF, and with my filter mods and an upgraded cap mine has 22uF (because i isolated the PSU reservoir with a filter, I upped the cap size).

You have an oscilloscope right? Track B+ line while playing audio. Is it a straight line? If it dips full 10 volts you still wouldn't see a change in performance, even when you might assume it's a bad thing. A bigger reservoir cap most likely doesn't do anything. I'm using B+ regulators with only a single 4.7uF poly cap these days, for example.

You mention the 1538, original G9 schemo called for the 1528.  I am assuming you are using the 1538.

I didn't realise you had changed it to LL1538 for the GIX51X. I'm using LL1528. Thing is, LL1538 has a completely amazing frequency response, all the way to 100khz! It performs even better than LL1528 looking at the datasheets.

There is nothing in the input stage that makes the bass cut you are seeing. I've been trying to explain this so many times already. What I don't get is that when you are measuring the whole signal path of G9 and see a bass cut, why have you automatically assumed it's the input transformer?

Please measure the input transformer in complete isolation (or with the line input pad) if you have trouble understanding what I'm saying.

 

[bruce0]

You have an oscilloscope right? Track B+ line while playing audio. Is it a straight line? If it dips full 10 volts you still wouldn't see a change in performance, even when you might assume it's a bad thing. A bigger reservoir cap most likely doesn't do anything. I'm using B+ regulators with only a single 4.7uF poly cap these days, for example.
 

You mention the 1538, original G9 schemo called for the 1528.  I am assuming you are using the 1538.

I didn't realise you had changed it to LL1538 for the GIX51X. I'm using LL1528. Thing is, LL1538 has a completely amazing frequency response, all the way to 100khz! It performs even better than LL1528 looking at the datasheets.

There is nothing in the input stage that makes the bass cut you are seeing. I've been trying to explain this so many times already. What I don't get is that when you are measuring the whole signal path of G9 and see a bass cut, why have you automatically assumed it's the input transformer?

Please measure the input transformer in complete isolation (or with the line input pad) if you have trouble understanding what I'm saying.

Measured 3V RMS dip approx, Only visible at low frequencies (below 200Hz), but I am not sure what the peak dip is...
Probably no effect.  Your 4.7uF poly sits next to the 100uF PSU reservoir in a G9?


Yes, I agree we don't seem to be communicating on this point.  It is probably my inexperience, or lack of clarity.

Clearly put: 

1) I noticed, measured and documented a difference in performance between the different input Transformers I have.  I am endeavoring to explain why, and/or improve the laggards.

I think you will agree the pre's measurements should not be different solely because of input transformer if they are not having an effect (EXCEPT for the inherent differences in their capabilities). And it makes little sense that an OEP should perform better than an LL1538 as is currently the case here.

Hmmm. Perhaps I should be testing the response without an output transformer.

2) I realize, recognize, agree with, acknowledge and appreciate the low frequency rolloff issues you have documented with respect to the SRPP output stage.  I appreciate your explanation.  I have not yet begun to approach those issues, and was planning on looking at it after I did #1 above.

I think I am hearing you clearly.  I understand that you suggest I address the "big obvious improvement" before I address the "little tweaky issue" that I am working on.  Thank you for your suggestion, and I will take a look at the "big" issue, but first I want to address the "little" issue.  I explain why this order below, if you care to know why.

I genuinely appreciate your comments, and apologize if I am causing you frustration (many of my teachers in the past had the same experience, and I think it is my fault).

Bruce

Explanation: Why am I working the input trafo issues first.  Because they should not be different in noise, distortion and performance from their respective specs and capabilities.  If I can make them perform to spec, I think I will have usable interesting colors in these preamps, and portable to boot. This is because I believe the Pre's will work for vocals and most strings with or without the extended low end, but I want the to have "color" not Jackson Pollack style distortion due to the input trafo setup.

bb

 

[Kingston]

Measured 3V RMS dip approx, Only visible at low frequencies (below 200Hz), but I am not sure what the peak dip is...
Probably no effect.  Your 4.7uF poly sits next to the 100uF PSU reservoir in a G9?

I meant I used a completely different B+ regulator with no reservoir, so to speak. I was designing this: http://www.groupdiy.com/drawing-board/b-regulator-with-variable-reference/msg651854/#msg651854

I wanted to test it in several places and it worked fine in G9 as well. There was a 100uF electrolytic in the non-regulated side next to the bridge rectifier. I didn't have a 10uF poly cap for the regulated side so I used 4.7uf.

What I'm really trying to say is that if your regulator is any good, there's no need for reservoir caps at all. If I understood it correctly from your other thread, the DC converter/regulator in the GIX51X model is pretty good.

Also just to make it clear, I enjoy reading about your meticulous tests. Just seemed you were probing the wrong end this time.

 

[bruce0]

I enjoy reading about your meticulous tests. Just seemed you were probing the wrong end this time.

Thanks!  Here are some more results from my having my head in the wrong end of the preamp 

Note: In the following tests I changed my names GIX1-4 are called Pre1-4 (sorry)

I put together a transformer in a box with selectable resistors in series with the output, and I measured the output impedance on my test rig so I could see what the impact of different input impedances were on the input transformer.

Here is the response of G951X with OEP in and OEP out with the zobel and load at -40dBu.

And same preamp with hot mic levels -20dBu

Here are the 4 preamps compared at typical mic levels and impedance.

Same preamp with DI and line level.  Line level is relatively unaffected by driving impedance (caption +16db is incorrect, it is really -26db)

Comparing DI to input transformer on Pre2 (Lundahl in - lundahl out) and Pre4 (Cinemag in - OEP out).

Note that the DI input appears in all cases (for a given output transformer) to extend the low end response (not of course as much as Kingston's approach) almost to the transformer's spec. So then I compared the response on all 4 preamps using the DI input, to compare the output transformers.  Oddly my Cinemag are showing up the Lundahls.  (Which I don't understand financially, I wonder if I wired them wrong)

Same test at higher levels into the DI

So I have to go figure out a couple of things, why are my OEP's behaving so much better than my Lundahls. 

I also note that the Cinemag performance being better than the other input transformers makes me think that it has to do with ratio.  I believe the Cinemag is wired as 1:10 step up and that may explain the better performance although I believe Jacob mentioned that this may result in hot mic's and loud sources distorting.

I need to compare these in use, because the response and the sound don't necessarily match.

And of course, eventually I have to get my head out of the wrong end of the preamp 8)

 

[dmnieto]

I am not sure if to resuscitate this thread... but if I were to cut some traces in the output transformer to wire the LL5402 as 4:1 (currently it is a 4:2 with both primaries and secondaries in series), would that allow to have the ECC82 drive the low frequencies better?

 

[bruce0]

Kingston has described how to extend low end response using a different tube, which I think has lower output impedance, but requires some "trace surgery" because the pinouts are different, and may require more power, see his notes.

I do believe the low frequency response of a tube output stage is improved if you increase the inductance of the output transformer primary and wiring it 4:1 would increase the inductance (but also lower the output level). I was wondering the same thing.  Have not done any test of that yet.

These things sound pretty good as designed, nothing thin about it.