DIY T4B, matching EL Panels and Cells

[DaveP]

Joe-electro,

An EL panel has quite a high blue component even though the CdS photocell has a response stretching from the green to the red, if it works ok then it shouldn't matter too much if it's off the optimum, a blue LED is very intense so that will compensate for any mis-match.

best
DaveP

 

[Ptownkid]

LOVE this discussion!

 

[SSLtech]

I was building my own from scratch so it was no problem to redesign anyway.  As I mentioned earlier, I think absolute matching between pairs is not so critical because a lot depends on the VU meter used.  I used a genuine VU meter from Sifam costing over $100 and I had to modify the associated bridge to calibrate the output to what it said on the meter.

I think I've always used proper VU meters, and I have a few of the REALLY nice Sifams sitting around here as of now. My current project is using a Beede... again!

The single biggest trouble which I've had with parallel LEDs is the turn-on point. If you wire ten LEDs in parallel, you'll find that some 'grab' current earlier than others. As a result, the threshold where the onset of GR takes place may be different from the moment at which the VU meter begins to indicate gain reduction, because they're not seeing the same amount of light at the early stage.

I really don't think the rise time of the LED/EL panel matters too much, we are only talking a few microseconds here, what matters more is the recovery time of the photocells and that has nothing to do with LED's or EL panels.

Definitely agree; for example the RELEASE time constant of the photocell utterly dwarfs any response time considerations of the light source, unless you're using something glacially slow like incandescent illumination. The differences between LEDs (when operated asymmetrically) and EL panels however can be quite significant, and its not to do with response speed. Steady-state  response of a looped "Aaaaaah" or bowed cello note shows marked differences in gain reduction when comparing symmetrical-versus-asymmetrical light emission. This is not time-specific, but waveform asymmetry-based.

Keith

 

[DaveP]

Hi Keith,

I take your point about the threshold of LED's being different but I think you'll find that all the differences are lost inside the slow fall time of the photocell, see this link.

According to the chart it takes ~400mS for the resistance to fall completely and about 200mS for it to even react.  I think the guy who posted this got the x scale wrong by an order of magnitude, but it does show that the response time of the photocell is slower than an LED or EL panel.
best
DaveP

 

[SSLtech]

I think I may be possibly misunderstanding you... or possibly you misunderstanding me...? Either way, response time is not what I'm thinking about.

I'm thinking not of transient differences, but of steady-state differences around the turn-on point.

When I build my first one, I used parallel diodes in "soixante-neuf", and observed that one direction both turned on at the same steady level, while in the other direction, only one was illuminating, until the drive crossed the SECONDARY threshold.

When I just wired all the LEDs in same-direction parallel, it was clear that some were begining there turn on before others, and when I say 'before' I mean not 'faster', but 'at a lower voltage'. Including steady-state.

This was probably about 10 years ago, certainly more than eight. -It may be that batch characteristics could be closer these days, but I've taken to heart that LEDs should never be trusted to act as 'identical twins'... more like 'fraternal twins', so to speak.

Not sure if I'm making myself clearer, or maybe it's me who's still misunderstanding you...? -In which case please forgive me... I'll figure it out eventually!

Oh, and I see the trimmer cap in your drawing now... C4 of course (duh! -Told you... sometimes it takes me while!) -Actually your layout os SO much clearer n terms of following the signal path that I was looking by the output transformer for it. -It really is a lovely drawing to follow.


Keith

 

[SSLtech]

Photos of my opto/LED coupling on my LA-4 by the way... I think these things cost about thirty cents each off the shelf:

 

[jensenmann]

I got the best results by putting the LED's behind a piece of white, translucent plastic to diffuse the light and leaving some space between the LED's and the plastic to get the light to disperse more evenly across the entire surface of the plastic.  I've also found that the bright blue LED's have the best dynamic range.

Joe

You can easily diffuse the light by sanding the LED with very fine emery paper or steel-wool.

 

[joe-electro]

You're misinterpreting the data here, which is understandable because you haven't been given any info about the test being shown here.  There are 4 samples at the beginning before the the light source turns on. This is to clearly show the "dark" state of the photocells at the beginning of the test.  The cells are then hit with a 250 millisecond pulse of light, which looks like 300 milliseconds on the graph because the samples are taken 0.1 second apart.  I've checked the accuracy of the tester and it's within ~1.5%.

This test was intended to be a rough interpretation of a classic engineering "delta" function, an infinitely short pulse that completely saturates the system under test so you can clearly see its impulse response.  I experimented with many different pulse durations before settling on 250 milliseconds as a "happy medium".  Shorter pulses don't excite the cells enough and longer pulses start to get into the cell's "memory effect", which affects the release times of the cells, which reduces the usefulness of the test.  The assumption is that in actual use nobody is going to be hitting the unit so hard that the light source is full on for more than a brief time during transients.  Since the LA-2A and LA-3A are both feedback designs this is a reasonable assumption.

Joe

Hi Keith,

I take your point about the threshold of LED's being different but I think you'll find that all the differences are lost inside the slow fall time of the photocell, see this link.

According to the chart it takes ~400mS for the resistance to fall completely and about 200mS for it to even react.  I think the guy who posted this got the x scale wrong by an order of magnitude, but it does show that the response time of the photocell is slower than an LED or EL panel.
best
DaveP

 

[DaveP]

Hi Joe-Electro,

Thanks for the explanation, I had no idea who had posted it, I've had it on my PC since I made my LA-2 3 years ago.  I certainly appreciate the thought that went into setting up the test.

I come at this from a slightly different angle.  I recognise that the NSL-5910 is absolutely essential to the character of the response and subsequent sound of the compressor.  As I am a phosphor guy, I simply looked for a green LED with the same characteristics as the EL panel.  The diode I found has the right rise and fall time, that is to say, very fast, it is the right colour, and it has enough output to drive the NSL-5910 to its lowest resistance.  For me it has ticked all the boxes.

best
DaveP

 

[joe-electro]

For me, as soon as the discussion turns to LEDs it's a whole different ball game.  I immediately think of going with a sidechain that is customized to the capabilities of LED's, primarily a DC drive voltage instead of AC, and that means we can build a full-featured sidechain with variable attack, release, ratio and other goodies.  Fortunately for us, someone has already done that and a PC board is already available. Thanks to Fred Forssell and the opto-compressor project he donated to the group!  I've been wanting to graft that circuit onto a tube audio path for some time, but it's still on the back burner.
Just thought I'd mention it for other DIY'ers that might have more time and motivation than I do at the moment.

Joe

 

[mjk]

"For me, as soon as the discussion turns to LEDs it's a whole different ball game.  I immediately think of going with a sidechain that is customized to the capabilities of LED's, primarily a DC drive voltage instead of AC..."

I would consider driving the LED with a current servo off the CV, which results in a more stable and dependable characteristic. Constant current PWM through a lookup ROM might be a good way to build your own curve. Of course, the timing is still down to LDR selection.

 

[mjk]

I found a short wavelength green LED with a fisheye lens. The first  idea was to grind a flat in the lens to butt against the LDR, but I found that alignment is not critical and a piece of hot-melt-shrink tube works great!

 

[solderspongebob]

May I chime in with a suggestion here: I've often wondered if there is a possibility of replacing the T4b with a microcontroller based circuit which would sample the rectified and scaled down voltage normally present across the EL panel, then create the desired resistance using FETs in substitution for the LDRs. The microcontroller could even mimic the recovery time of an ideal LDR.

Maybe if the voltages/currents allow, then digital potentiometer chips could be used in place of the FETs which would give superb accuracy.

If that could work, then next would be adjustable threshold and recovery curves, stereo linking etc etc.

Just a thought.

 

[joe-electro]

It's definitely possible, although I had thought of using a DSP development board instead of a microcontroller, to get better resolution.  I think you would need at least 16 bit resolution to avoid 'zipper effect'.  The math would be fairly complex because you're dealing with exponential curves with hysteresis superimposed to mimic the EL panel, and FET's are very nonlinear devices with a 1/2 power transfer function (if memory serves).  Getting that kind of system to work would be a fun project indeed!

Joe

May I chime in with a suggestion here: I've often wondered if there is a possibility of replacing the T4b with a microcontroller based circuit which would sample the rectified and scaled down voltage normally present across the EL panel, then create the desired resistance using FETs in substitution for the LDRs. The microcontroller could even mimic the recovery time of an ideal LDR.

Maybe if the voltages/currents allow, then digital potentiometer chips could be used in place of the FETs which would give superb accuracy.

If that could work, then next would be adjustable threshold and recovery curves, stereo linking etc etc.

Just a thought.

 

[MeToo2]

I'm going to ask a heretical question: what does it matter if the cells don't match?

The LA-2a is (mainly) used on mono sources if I'm correct (vocals, bass).

I can imagine the dark resistance has to be high (say >150K), so that there's no compression on no side chain signal.

I can imagine the light resistance has to be low to give decent compression when hit hard and achieve good dynamic range (say 150 ohms) to get ±60dB.

Since the input signal can be varied louder or softer, I can imagine the absolute level of compression for a given input signal isn't that important.

I can imagine that the attack time isn't so important (I don't see a lot of lag on the graphs)

I can imagine the release time constants are very important to get a distinctive sound = release time.

I can imagine that the shape of the compression curve for a constant source level is very important to get a distinctive sound.

But how "good" does it have to be to sound in the ball park for a poor mans DIY version tube based opto compressor that won't even use the same transformers etc. ?

To be honest, I wouldn't care if the GR meter didn't track that accurately. That can always be tweaked with an offset and scaling to be "right" at two points on a straight line e.g. 0dB and -6dB compression.

I spent half a day matching two quads of FETs for the stereo G1176 and that was time well spent (it only cost 30 euros or so for a 100 FETs) and the stereo pairs and GR matched within 1dB across a big range of compression.

But if I'm going to have to buy 20 photo cells to get one pair to match, or say 50 for 4 cells, I probably may as well buy 1 or 2  commercial T4b replicas.

So what's the expected ratio for 2 usable GR cells that sound OK for 2 units, + 2 GR meter cells that track somewhere near?

 

[joe-electro]

A) If you're using your units strictly on mono sources then unit-to-unit matching of T4B's is not as important

B) The dark resistance must be MUCH higher, over 5 megohms for the unit to have negligible affect on the signal level

C) Around 1k at full brightness is typical

D) Input level to the unit must be high enough to drive the sidechain amplifier to the desired level of gain reduction

E) Probably not. Most of the light source/photocell combinations in use have very fast attack times

F) Yes, this is probably the most critical characteristic

G) If you're referring to the attenuation curve relative to the input level you are correct.

H)  There are three types of matching I do.  First, the two cells in a T4B unit must match each other exactly.  While you are correct that it is possible to have a meter read correctly at two points on the scale (e.g. 0 dB and -6 dB) under static conditions, the real benefit of the meter is being able to tell what's happening *dynamically* while the music is playing, especially when the unit is in the release phase. If you want the meter to really tell you what's happening to the signal the photocells must match almost exactly.

Second, matched pairs of T4B's must match almost exactly.  This keeps the stereo image from wandering left to right or vice versa.  There are plenty of instances where a stereo pair get used, on the mix bus, on drum overheads, on a stereo instrument.

Third, the cells must match the ideal attenuation and release curves.  This is what makes an LA-2A or LA-3A sound the way they're supposed to.  If the cells are too fast or too slow the whole unit doesn't sound right.


I) If you trust your ears you don't need a meter at all. Personally, I want the meter to be as accurate as possible, which is what I demand of everything in my signal chain.  Expecially during recalls, I want to know that I'm getting repeatable results. But I want my opto compressors to sound right, and that means having the proper photocells in them.

J) Right now the ratio is about 10:1.  The thing is, the photocells that are being shipped now are all over the place - very inconsistent - so you need a large pool to draw from in order to get a decent match.  It's like trying to draw a pair of aces from a deck.  If you only have four cards to draw from, your odds are much worse than if you have all 52.

If all you want is a T4B that sounds pretty good for use in your personal project studio the standards are probably way lower.  Over the years my standards have gotten higher and higher as I've refined the process and reaped the rewards of my labor.

Bottom line: you've got to have good cells in your T4B if you want your unit to sound like the real thing.


Joe

I'm going to ask a heretical question: what does it matter if the cells don't match?

A) The LA-2a is (mainly) used on mono sources if I'm correct (vocals, bass).

B) I can imagine the dark resistance has to be high (say >150K), so that there's no compression on no side chain signal.

C) I can imagine the light resistance has to be low to give decent compression when hit hard and achieve good dynamic range (say 150 ohms) to get ±60dB.

D) Since the input signal can be varied louder or softer, I can imagine the absolute level of compression for a given input signal isn't that important.

E) I can imagine that the attack time isn't so important (I don't see a lot of lag on the graphs)

F) I can imagine the release time constants are very important to get a distinctive sound = release time.

G)I can imagine that the shape of the compression curve for a constant source level is very important to get a distinctive sound.

H) But how "good" does it have to be to sound in the ball park for a poor mans DIY version tube based opto compressor that won't even use the same transformers etc. ?

I) To be honest, I wouldn't care if the GR meter didn't track that accurately. That can always be tweaked with an offset and scaling to be "right" at two points on a straight line e.g. 0dB and -6dB compression.

I spent half a day matching two quads of FETs for the stereo G1176 and that was time well spent (it only cost 30 euros or so for a 100 FETs) and the stereo pairs and GR matched within 1dB across a big range of compression.

But if I'm going to have to buy 20 photo cells to get one pair to match, or say 50 for 4 cells, I probably may as well buy 1 or 2  commercial T4b replicas.

J) So what's the expected ratio for 2 usable GR cells that sound OK for 2 units, + 2 GR meter cells that track somewhere near?

 

[MeToo2]

J) Right now the ratio is about 10:1.  The thing is, the photocells that are being shipped now are all over the place - very inconsistent - so you need a large pool to draw from in order to get a decent match.  It's like trying to draw a pair of aces from a deck.  If you only have four cards to draw from, your odds are much worse than if you have all 52.

If all you want is a T4B that sounds pretty good for use in your personal project studio the standards are probably way lower.  Over the years my standards have gotten higher and higher as I've refined the process and reaped the rewards of my labor.

Bottom line: you've got to have good cells in your T4B if you want your unit to sound like the real thing.


Joe

Thanks for the reply. Not questioning your standards at all. On the contrary, I'm questioning myself on whether I can recreate anything like the mojo for a reasonable DIY cost compared to just buying a ready made OEM component. I know these things are difficult to get hold of due to RoHS etc.

Hmmm. Wasn't really considering stereo myself. Just seemed too difficult. Possible out would be to go for a mid/side codec around your compressors to try avoid left/right image shift.

If the ratio is indeed 10:1, then for my 2 T4b's I'd need 40 cells. With minimum orders that's probably more like 50 or even 100. That's going to be 50*$3 before I mess around too much. But please don't put your prices up on my account For that sort of money you're anyway getting near original UREI part prices if you want to go the whole hog.

Does the colour difference of the cell material suggest that the manufacturing process has changed significantly for the Silonex?

My guess is that the shape of the curves will be largely material dependent. So if that has changed significantly, you're not really buying anything like the original part.

Maybe a group buy might be in order. RS go down in price at 100 and 250.

Or else a search for another cheaper photocell if the NSL5910 really are so all-over the place anyway e.g. VT43N1, VT83N1, or VT80N1.

There must have been very many who have wandered this path before. I know people have had success with Vactrol vtl5c2/2 in the past in opto compressors.

Did anyone already contact Silonex to confirm a material change? And perhaps to ask for recommendations for alternatives to the NSL-5910?

Lots of questions I know.

 

[joe-electro]

If anybody's interested, I just posted this video on YouTube, comparing a Kenetek, Universal Audio reissue, and rebuilt vintage UREI T4B unit using the Kenetek T4B test fixture:

http://www.youtube.com/watch?v=HB86Mw4mZLw

Thanks,

Joe

 

[MicDaddy]

Joe,

Thanks for sharing that video! 

 

[DaveP]

Hi Joe,

Really good Youtube video, well done.

You need a new Avatar, you're really a regular guy, not weird at all!

Regarding the EL panels:  I can fill in a bit of history here.  The development work was done in the early 60's by many companies, mainly in the US.  They even talked about making giant wall panels for general lighting.  Apart from the child's nightlight application, they also hoped to use them for car instrument backlighting; the problem is that they need ~100Vac to work so that was a problem with only a 12V supply.  They lapsed into obscurity until the mobile phone industry required a backlight for their displays in the 90's, electronics had solved the high voltage problem by then.

Sylvania was the only company still making the stuff, but by then they had developed a waterproof coating which extended the life considerably.  No-one else was able to achieve the same level of brightness as the Sylvania phosphor, I know because my company tried and failed!  I think all the panels available now use Sylvania phosphor and they have cornered the market, (min order used to be $2000, about a pallet load) and their material is certainly better than that available in the 60's due to the continuous development over the last 50 years.

Although they start working at ~100Vac they increase in brightness up to 400Hz and start to tail off again past 1kHz, I think this is why there is a frequency shaping circuit between the 12AX7 and the 6AQ5A, the frequency response is thus the reciprocal or opposite of the EL panels response to frequency.  I left it in on my Green LED circuit but maybe that was a mistake thinking about it now.  I'd be interested to hear what others think on the subject.
best
DaveP