etheory - a bunch of projects in progress

[tv]

All appear to have rather similar specs for power handling, hfe and potentially noise.

That's the point. So you can "vibe" the best candidate for that particular circuit... bc140 also comes in diff. Hfe grades.

Although I have my doubts that the noise figure will matter much in the output drivers, especially when slamming drum loops... but who knows, somebody may occasionally want to record some very delicate, miniscule tones, coming out of that particular "endstufe"!

These will all look very cool when soldered on the board, and THIS is what truly matters.

 

[Triode-Joe]

I'd be happy to provide some mod ideas and how to do them. I'll make sure to post that here + some proper diagrams over the new few weeks. But that should be no issue, aside from snipping wires in your compressor a little bit.

Thank you for the ideas and the notes on the U274 schematic. From the audio clips, your design sounds far cleaner on compression and I would think the NE5532 buffer presents a more consistent load (impedance) than the input transformer. I think I'll limit my work to mods of the daughter board of the original. At some point, it would be great to hear a "side by side" comparison of your design and a U274.

If you know how many compressors / limiters /EQs you have, than you don't have enough! 

TJ

 

[etheory]

No worries Triode-Joe! Hope someone tries the mods with their old U274's and gets some success!

At some point, it would be great to hear a "side by side" comparison of your design and a U274.

Me too. I still have to re-cap and re-box my originals. But I feel that will take longer than finishing my "clone" - haha.

 

[etheory]

e274 Compressor:

Test time before I finalize everything (which I am hoping to do this weekend)....

Firstly, no compression, just the end to end audio path, sitting on a breadboard, right next to a power point.
i.e. worst possible test case....

Testing three possible amplifier transistor combinations using RMAA (identical test conditions), impedance balanced output, electronically de-balanced front end:

20Hz to 20kHz


BC109B + 2N2219A

Frequency Response (multitone), dB:                          +0.06, -0.28
Noise Level, dBA:                                                          -70.4
Dynamic Range, dBA:                                                      70.8
Total Harmonic Distortion (THD) %:                                0.016
Intermodulation Distortion + Noise %:                            0.203
Intermodulation Distortion + Noise - swept (freqs) %:    0.122
Frequency Response (swept sine) dB:                          +0.1, -0.2

BC109C + 2N2219A

Frequency Response (multitone), dB:                          +0.06, -0.19 (goes to +0.05, -0.19 with 0.1uF supply decoupling)
Noise Level, dBA:                                                          -71.0 (goes to -75.9dB with 0.1uF supply decoupling + 100uF)
Dynamic Range, dBA:                                                      70.8 (goes to 77.0dB with 0.1uF supply decoupling + 100uF)
Total Harmonic Distortion (THD) %:                                0.014 (goes to 0.012 with 0.1uF supply decoupling + 100uF)
Intermodulation Distortion + Noise %:                            0.217 (goes to 0.164 with 0.1uF supply decoupling + 100uF)
Intermodulation Distortion + Noise - swept (freqs) %:    0.117 (goes to 0.106 with 0.1uF supply decoupling + 100uF)
Frequency Response (swept sine) dB:                          +0.1, -0.2

BC550C + 2N2219A

Frequency Response (multitone), dB:                          +0.07, -0.16
Noise Level, dBA:                                                          -71.1
Dynamic Range, dBA:                                                      71.5
Total Harmonic Distortion (THD) %:                                0.017
Intermodulation Distortion + Noise %:                            0.216
Intermodulation Distortion + Noise - swept (freqs) %:    0.110
Frequency Response (swept sine) dB:                          +0.1, -0.2

Tested with RME Fireface 800

- now, happy with the quality of the amplifier that's driving this tiny little compressor, it's time to quickly mess with some better PSU filtering / decoupling, then on to plotting some compression curves....

 

[tv]

If you're still in "mod" vibe, you could try to bootstrap the amp, like this.

For the "sonics", it would bring it more "upfront" - but that may not be what's really wanted in this particular design ... ymmv.

(note - not tested with this particular amp)

 

[etheory]

Thanks TV! I'll have to try that!

OK, here are 3 curves.

Using the UC3611N matched diode array for the compression element (gray dots).
1.) The 1-1 reference curve measured with compression disabled - adjusted for unity gain end to end.
2.) The same process, at the same volume, with compression enabled. (red dots)
3.) At 0dB input volume, the compressor is adjusted out to JUST before clipping. Then the process repeated. (blue dots)

Seems the compressor with this element does just under 3:1 (about 2.6:1), is that right? The slope of the line is 1/3 at full compression, but it's in a linear scale. Do I have to measure that on a log scale? It certainly SOUNDS like harder limiting than that, but maybe it's not....

It's pretty clear I'd need about 20dB of further output gain (17 to be closer to precise) with this attenuator element in the circuit.

Next I'll measure the 1N4148 and 5.1V Zeners to see how they perform and what sort of gain I'd need for them.

 

[etheory]

I finally just realized why the value of C3 sitting near the diode bridge is the value it is.
I raised it to 22uF in my circuit thinking it wouldn't make much difference, initially thinking it was just to smooth out the negative feedback and Ts1 bias signal, but then, after changing it to be the same ratio (4.7:1) as it was in the original circuit, a wonderful thing started to happen.

The weird bias + the 10uF C3 capacitor, removes a LOT of the CV feed-through in that circuit in a very clever way.

I reduced C1, C2 to 10uF for a much faster attack time, which means that C3 I'll set to 2.2uF to get a close-ish ratio to the original (which used 47uF to 10uF).

What happens is that C3 and R13 and R14 form a low pass filter that smooths the compressed signal flowing through the emitter of Ts2, along with providing negative feedback via the input transformer to Ts1, that bias and negative feedback line also contains an inverted filtered signal, consisting mostly of the CV component of the signal, and hence due to the negative feedback, partially cancels it out of the signal path.

When I first found this I couldn't believe it, it's genius. This U274 circuit, even though it's basically just this one (with a few modifications):

is so subtly clever. Those Siemens Sitral guys were on the money for the parts count.

There is now almost no CV feed-through at all. Pretty cool stuff!

If you remove that capacitor entirely, there is no CV feed-through at all, but it comes at the cost of attack speed, which drops significantly.

 

[etheory]

Here are 3 recordings, one for each diode configuration (No effort was made to match levels just yet, that'll come later with the makeup gain stage):

UC3611N:
http://www.evolutionarytheory.com/wp-content/uploads/2013/04/e274_Limiter_Recording_UC3611N.mp3

5.1V Zener:
(Coming soon)

1N4148:
http://www.evolutionarytheory.com/wp-content/uploads/2013/04/e274_Limiter_Recording_1N4148.mp3

(edit - I might redo these once more in a single run to ensure consistency - these were done over a few consecutive days.

 

[tv]

I finally just realized why the value of C3 sitting near the diode bridge is the value it is.
....
There is now almost no CV feed-through at all. Pretty cool stuff!
....

Hmm... yes and no.

The way it looks to me the C3 is connected via the node "16" directly to ground.

What this arrangement reminds me of, is "remote ground sensing". Ie, the way this is connected, the Ts2 transistors is "listening" to the shortest path to the variable resistance realized via diode bridge, C1 and C2 and C4.

If you look at it differentially, the Ts2 emitter is like a "-" input, and it is connected to the "cold" point of said diode bridge circuit, listening to the audio signal across the bridge. Naturally, audio levels here should be pretty low, so any charge/discharge current flowing thru diodes and C1/C2, (via C5) would result in some "dirty" voltage being injected into the "ground" node, via traces and component leads etc. I assume the general audio AC levels here are small, so this "dirty" ground AC would be added to the audio signal and promptly amplified.

My guesstimate is that connecting the C3 directly to the diode-bridge circuit "senses" the audio signal normally, but the "dirty" component differentially. Similar tricks (where it was important to connect the gain FB network to the "right" ground node) were actually quite common in power amps, preamps etc.

But hey, I may be wrong.

For giggles, you could try relocating the C3 right across the R14 leads and see (actually "hear") what happens...

 

[etheory]

I doubt you are wrong. My circuit analysis skills are level: newborn whereas you clearly have a much better idea of what you are talking about.

But what I have observed directly (on a scope, whilst prodding around the circuit) is the profound effect it has on CV rejection.

No doubt it also affects ground current injection (it seems this deceptively simple circuit uses a lot of points in the circuit for multiple reasons).
The point you make about differential rejection is definitely true.
If I remove that cap it's clean as a whistle, but the attack time is hampered by the amp "trying to push back to the input signal", as far as I can tell, and that differential circuit "removing the compression".
It seems the value of that cap is optimized to reduce the speed of that feedback to a level where transients can come through. I think....

At least, when I change the cap to be bigger, I get far more CV coming through, and if I make it smaller, then the attack time is severely hampered.
It could be the quietness of the circuit is, however, due to that "dirt removal" you are speaking of. I might try relocating that cap to different points on the breadboard to see if it's locality to "dirty return locations" effects noise performance.

Thanks for the tips!

 

[evilcat]

Here are 3 recordings, one for each diode configuration (No effort was made to match levels just yet, that'll come later with the makeup gain stage):

UC3611N:
http://www.evolutionarytheory.com/wp-content/uploads/2013/04/e274_Limiter_Recording_UC3611N.mp3

5.1V Zener:
(Coming soon)

1N4148:
http://www.evolutionarytheory.com/wp-content/uploads/2013/04/e274_Limiter_Recording_1N4148.mp3

Cooool, sounds good man!!! I like the 1N4148 way more than the UC3611N, it seems to give a more "coherent" compression, if it means something , more pleasant. With the UC3611N it seems that most of the time only the kick drum drive the compression...
Anyway, good job man. So I'm a bit lost : what controls will there be ? ratio, attack, release ? threshold & make up gain of course...

best,

ben

 

[tv]

No point in overanalizing, but I'd like to point out that the C3 will directly determine the erm.. "loop gain" of the amp, which is going to manifest as "lack of compression" - or a much slower, softer reaction - when the C3 is removed or insufficient, because of a feed-back nature of the circuit.

If the C3 is too big, it will move the amp's corner frequency lower, and thus making it more sensitive to CV feed-through.

Theorizing: If the C3 is smaller, this would theoretically manifest in higher frequencies being squashed with more of a "hard knee", and lower frequencies with more of a "soft knee".

So, you could start to think of it as a potential "tone control" tweak-thing. For example, wiring a 1uF/10k pot. combo in SERIES with the C3 (as a sort of a faux HPF) and listening for the sweet spot (if any) and going from there to the optimal component values - "for the ultimate listening pleasure".


But hey,

 

[tv]

Here's what I meant in my above comments..

Perhaps it will work as I envisioned, perhaps I was wrong.

 

[evilcat]

tone control on a compressor? I like that. Kind of Chandler-ish control!!!

 

[Kingston]

Here are 3 recordings, one for each diode configuration (No effort was made to match levels just yet, that'll come later with the makeup gain stage)

Thanks for posting these.

Have you scoped or simulated the effects of these diodes much?

The only difference, besides forward voltage drop, would be the reverse breakdown voltage spike. And that's so fast it shouldn't have any audible effect.

The first diode sounds like it's distorting more than 1N4148. Might be as simple as a level difference from forward voltage drop of the diodes. And since it's in a CV loop call it "threshold".

I would recommend simulating this diode configuration in isolation with real world input level. No point in complicating this with listening tests. You could then perhaps move into active biasing tricks where you can play with the forward voltage drop, no longer limited to diode selection.

 

[etheory]

Thanks for posting these.

You are more than welcome.

Have you scoped or simulated the effects of these diodes much?

In a lot of detail, but simulation difference is nothing like what is occurring in the real world situation.
The simulations don't show the type of distortion you get when the diodes begin to conduct out of their ohmic region (where the compressor starts generating nasty distortion, which starts at a higher signal level for a silicon than zener/schottky/germanium, hence the 1N4148 can take more input signal more cleanly).

The only difference, besides forward voltage drop, would be the reverse breakdown voltage spike. And that's so fast it shouldn't have any audible effect.

The first diode sounds like it's distorting more than 1N4148. Might be as simple as a level difference from forward voltage drop of the diodes. And since it's in a CV loop call it "threshold".

Well there is one thing that occurs that means the circuit can't be the same for different diodes.
The schottky basically conducts for "all" CV levels - it is so sensitive, that unless I have at LEAST 6.2K in the feedback loop to reduce the current in the feedback loop, the whole amp oscillates at high frequency around the CV feedback loop, due to the direct path and the lack of resistance of the Schottky array.
The 1N4148 doesn't do this due to it's higher signal level requirement to start conducting in the ohmic region. In fact, you don't need any resistance at all in the feedback loop with it, and it's higher signal level handling also makes a difference.

I would recommend simulating this diode configuration in isolation with real world input level. No point in complicating this with listening tests. You could then perhaps move into active biasing tricks where you can play with the forward voltage drop, no longer limited to diode selection.

Well, at this point, I think for this particular circuit I've messed with it about as much as I can be bothered messing with it ;-)

The 1N4148 is also giving me a much higher compression ratio also, which is great, and doesn't suffer from any oscillation issues like the lower voltage diodes can, so it seems like a more sure-fire bet to put in a kit for other people, as it's much more likely to give consistent behavior. However you could also put a 10K trimmer in the CV feedback loop to trim out any high frequency oscillation that might occur due to your particular choice of diode.

So I think I'll clean up the design now, re-do the circuit layout and build another proto over the weekend.

Thanks so much everyone for your help. I'm not ignoring the feedback, it's more that I think it's time to wrap this one up and move on!

However, I will be revisiting the diode topology very soon in a couple of different ways.

Controls will be:

1.) Input volume (after debalancing NE5532)
2.) Output gain (NE5532 balancing stage)
3.) Passive HPF control (12 position rotary switch)
4.) Ratio (12 position rotary switch)
5.) Release (12 position rotary switch)
6.) Slow and Fast timing (no separate attack control) - 2 or 3 position switch, up to you.
7.) Compressor On/Off (leaving you with just an amp with about 48dB gain max, end to end).

 

[etheory]

Here are the "final" compression curves in Blue for the UC3611N and in Orange/Red for the 1N4148.
This is with the "ratio control" set to "max".

Light color in each tone means it was created with the amp still at 1:1 volume (0dB in 0dB out) and then the compressor engaged at that point.
The darker color in each tone is the highest signal level that could be achieved before 0dB JUST started clipping.

Grey is the amplifier reference with the compressor disengaged.

All curves taken at 1kHz, and 0dB representing the max input level (+4dBu) of my soundcard.

 

[Kingston]

Well, at this point, I think for this particular circuit I've messed with it about as much as I can be bothered messing with it

I didn't quite realise the level of detail you had simulated and tested these. I've experimented with different diodes only in simple full wave rectifiers in compressors. Had no idea how delicate it can get with different topology. I now slither back to the hole whence I came.

What a great thread and highly informative data.

 

[etheory]

I didn't quite realise the level of detail you had simulated and tested these. I've experimented with different diodes only in simple full wave rectifiers in compressors. Had no idea how delicate it can get with different topology. I now slither back to the hole whence I came.

Sorry Kingston! I didn't mean it like that. It's merely a mistake of apparent timing - I was simply getting ready to wrap up the project anyway as I'd already spent a couple of months on it.
On the one hand I would love to continue refining, but I've already invested too much time and am happy enough with the current configuration.

I also have to leave some time to disseminate my findings into my next few projects ;-)

After this one is wrapped up and the boards are made and for sale, I'm moving on to a modern version of the U273. I once read someone say a Telefunken employee referred to it as a boat anchor, so I might see if I can turn it into a Ferrari boat anchor ;-)

After that I already have my "ultimate diode topology" compressor in the works, which draws upon and refines all of these ideas into a more modern/accurate/low noise version.

But all that later.... First I godda get these PCBs made and see if anyone wants to get their hands on a dirty old compressor design with a lot of character and very little accuracy! 8)

(FYI, the compressor curves were produced using OpenOffice Calc and the Operator synth in Ableton Live set to a 1kHz sinewave. I just set the volume control on Operator to the plotted input volume, then read off the output volume through the compressor. You have to wait a little bit for the attack to come through, so you have to hit the peak reset on the meter a few times between each reading and wait a couple of seconds between plot points, but then you just read off the incoming level and put it in the table. Then I just plotted the result with the Calc graph tool.)

 

[evilcat]

Controls will be:

1.) Input volume (after debalancing NE5532)
2.) Output gain (NE5532 balancing stage)
3.) Passive HPF control (12 position rotary switch)
4.) Ratio (12 position rotary switch)
5.) Release (12 position rotary switch)
6.) Slow and Fast timing (no separate attack control) - 2 or 3 position switch, up to you.
7.) Compressor On/Off (leaving you with just an amp with about 48dB gain max, end to end).

12 pos HPF filter, isnt it too much? 3/4 pos would be already cool... my 2 cents!
No iron there, it should be a cheap build, isn't it?
What about PSU, will there be an on-board power section?
Could you linked 2 units for stereo?

best,

ben