How to build a simple clip detector ?

[Neeno]

Hi guys...
I'm looking for a simple solution to build a clip detector for my Neebe 1272.

Requisites
- It must be simple
- I'm going to build it on veroboard
- Just one led needed, that turn on at (-3dB ???)
- It should work @ +24V

I don't know where it should be connected and if it needs to be buffered..
Any suggestion it's really appreciated.

 

[keefaz]

There is a Peak Detector from Rod Elliott's projects at :
http://sound.westhost.com/project30a.htm
(near the bottom of the page)

 

[Neeno]

This is exactly what i was looking for !
Thank you very much Keefaz !

Cheers :guinness:

[quote author="keefaz"]There is a Peak Detector from Rod Elliott's projects at :
http://sound.westhost.com/project30a.htm
(near the bottom of the page)[/quote]

 

[JohnRoberts]

I second the use of a proper comparator (like 339). Using an opamp as a comparator may require dealing with input bias currents.

JR

 

[bcarso]

Yeah I use a lot of 339/393 parts too for various chores like these. The wire-AND capability of the OC outputs often simplifies things.

Once in a great while the tendency for a little oscillation burst at the trip point requires attention, but for most apps it can be ignored.

 

[bcarso]

One caveat for those using simulators and not getting to the bench very often: the application of a bit of positive feedback to these parts, to reduce or elminate chatter at thresholds, throws Spice in my version at least into fits. The reality on the breadboard is usually quite stable and benign by comparison.

OTOH the simulator will generally not reveal the trip-point-vicinity oscillations unless you add parasitics to the model.

 

[JohnRoberts]

[quote author="bcarso"]One caveat for those using simulators and not getting to the bench very often: the application of a bit of positive feedback to these parts, to reduce or elminate chatter at thresholds, throws Spice in my version at least into fits. The reality on the breadboard is usually quite stable and benign by comparison.

OTOH the simulator will generally not reveal the trip-point-vicinity oscillations unless you add parasitics to the model.[/quote]

I never really got comfortable with circuit simulations, it seemed like there was a catch 22 where you had to know what you didn't know to program the simulation to account for what you didn't know..? How's that for circular mumbo jumbo logic?

I did write some crude software to help me design oddball multi-pole cascaded filter sections back in the bad old days ('70s) having to design anti-alias, anti-imaging, and reconstruction filters around BBD and CCD delay lines (with HF pre-/de-emphasis included). I used the tab function to print crude frequency response plots on an old dot matrix printer, but I'm wandering off topic.

My trouble with circuit sims is I was always running into subtle component quirks like with the good old cheap 3086 transistor array. While a better sim may include the intrinsic substrate diode, I've seen really obscure things like interactions between adjacent devices from heavy current in one transistor affecting the one next to it in strange ways. Of course it only shows up in production because the working prototype allocated the transistors differently. Now maybe I was a little off sheet wrt to the devices typical applications but welcome to my world.

It always made sense after you figured out what was what. But then there's the short list that never made sense. :?: Perhaps I'm just showing some old dinosaur butt.. I really love computers for PCB and mechanical design. You can make stuff now where the first cut actually fits together.

JR

 

[bcarso]

I used to be almost as sim-averse as Pease, who is shown lining birdcages with printouts and hurling machines from rooftops.

But then my tech showed me how a colleague's filter banks all oscillated (and it wasn't obvious from inspection, as they were Friend-Delyannis topologies and a bitch to design). Turned out he'd made a sign error and had a bunch of poles in the right half plane (and boy was he embarrassed). After a bit I dipped my toe in using a free version of creaky old circuitmaker. I liked it immediately because I didn't have to list nodes and such---it extracted the necessary data from the schematic.

But you must know how things work to begin with or you will be hopelessly misled. That's part of the basis for Bob Pease's disdain---he would get these phone calls saying "your part oscillates!" and after a bit find out that the person hadn't actually built anything yet, and was a relatively poor user of a possibly flawed simulator.

I continue to be perplexed by what does and does not converge, and why a tiny change in a value will allow the program to proceed. Obviously this would be devastating for someone with no feel for what is really happening ("You must never use a 1.00k resistor here, only a 1.001k, or everything will blow up").

The fourier analysis routine I've come to love, while taking its predictions with a very healthy amount of salt. I've shoe-horned it into doing IM which has been most illuminating.

The most glaring omission I can see for the program, and something common to most of the affordable simulators, is an explicit signal-induced self-heating analysis capability. I worked out how to insert this with a bunch of math function modules and so forth but it is a lot of work.

 

[bcarso]

JohnRoberts wrote: "My trouble with circuit sims is I was always running into subtle component quirks like with the good old cheap 3086 transistor array. While a better sim may include the intrinsic substrate diode, I've seen really obscure things like interactions between adjacent devices from heavy current in one transistor affecting the one next to it in strange ways. Of course it only shows up in production because the working prototype allocated the transistors differently. Now maybe I was a little off sheet wrt to the devices typical applications but welcome to my world. "

I do miss that array. It was truly the transistor array of a thousand and one uses. Nasty popcorn noise in its initial process version though, but cheap and usually very well matched parts, even when you didn't pay for the offset voltage screening of the 3046. But I was always trying to cheat a bit and get away with using that Q that had the emitter tied to the substrate (pin 13 (?)), and it can get one into trouble. And they did have horrible current-crowding about 10mA or so iirc. Do you remember the ceramic DIP with the gold cap frit-seal, the CA3045? Those were beautiful, although it's true I didn't get out much those days.

STMicro made some 3086/3046 parts for a while. They just copied the original RCA datasheet though,, and had I not tested them I would have assumed that they simply duplicated the process. But actually they were WAY better parts---no popcorn noise, much higher beta, and probably still nice and fast. But then they discontinued them :sad:

Then Harris teased us with some superficially-sounding-similar arrays of superfast N's and P's. I don't know if those are still available or not (I think they had an HFA prefix). Blinding speed but low voltage.

 

[JohnRoberts]

[quote author="bcarso"]JohnRoberts wrote: "My trouble with circuit sims is I was always running into subtle component quirks like with the good old cheap 3086 transistor array. While a better sim may include the intrinsic substrate diode, I've seen really obscure things like interactions between adjacent devices from heavy current in one transistor affecting the one next to it in strange ways. Of course it only shows up in production because the working prototype allocated the transistors differently. Now maybe I was a little off sheet wrt to the devices typical applications but welcome to my world. "

I do miss that array. It was truly the transistor array of a thousand and one uses. Nasty popcorn noise in its initial process version though, but cheap and usually very well matched parts, even when you didn't pay for the offset voltage screening of the 3046. But I was always trying to cheat a bit and get away with using that Q that had the emitter tied to the substrate (pin 13 (?)), and it can get one into trouble. And they did have horrible current-crowding about 10mA or so iirc. Do you remember the ceramic DIP with the gold cap frit-seal, the CA3045? Those were beautiful, although it's true I didn't get out much those days.

STMicro made some 3086/3046 parts for a while. They just copied the original RCA datasheet though,, and had I not tested them I would have assumed that they simply duplicated the process. But actually they were WAY better parts---no popcorn noise, much higher beta, and probably still nice and fast. But then they discontinued them :sad:

Then Harris teased us with some superficially-sounding-similar arrays of superfast N's and P's. I don't know if those are still available or not (I think they had an HFA prefix). Blinding speed but low voltage.[/quote]

I think the last time I used one (3086) was maybe 10-12 years ago in an automatic mixer.. precision rectifiers and log conversions on every input to do the dB gain sharing math. The product has probably been redesigned once or twice since I left, wonder what's in there now? Today a junior engineer would just grab for a dbx chip that does all the heavy lifting or maybe just do it in the digital domain if cheap enough. The dbx chip sets were in the system and active even back then (I used their VCA) so that's my guess.

I remember the better series transistor arrays. IIRC there were some with PNPs in the mix but I could never justify using them on price Maybe when all the old parts I like go obsolete I'll just stop designing. :sad:

These days I can't ignore how much you can do with one of these swiss army microprocessors. Multiple A/Ds, dual PWM that can be used as crude D/A, flash memory built in, etc... Products can be far more clever and do tricks that simple analog circuits never could.

Tempus fugit.

JR

 

[clintrubber]

[quote author="Neeno"]Hi guys...
I'm looking for a simple solution to build a clip detector for my Neebe 1272.

Requisites
- It must be simple
- I'm going to build it on veroboard
- Just one led needed, that turn on at (-3dB ???)
- It should work @ +24V

I don't know where it should be connected and if it needs to be buffered..
Any suggestion it's really appreciated.[/quote]

This will be about the simplest way to do it:

(ignore R5 & feed signal at the top of R7)

Needs some upside-downing for the +24V (& use a NPN) & redimensioning for the exact clip or signalling level, but you'll get the idea.

And of course it doesn't do peaks both ways, but as 'simple' was on the top of your list...

Since this circuit is so easy you could even put a few more in, to monitor where things clip in the signal path. The more elaborate methods can do that as well of course, these can also be made to monitor more signal-points. But these often keep a 'single backend' so with one LED you know there's something clipping but just not where exactly. FWIW...

Bye,

Peter

 

[[silent:arts]]

for +24V, look at Neumann and Telefunken:
Neumann U411A - Peak Indicator for the V476 Mic Pre
Telefunken V6761 - Peak Indicator for the V676b Mic Pre

 

[PRR]

> I never really got comfortable with circuit simulations, it seemed like there was a catch 22 where you had to know what you didn't know to program the simulation to account for what you didn't know..?

I like to say: Design it on a cocktail napkin, and prove it will work, FIRST!

Then let the idiot assistant derive an 8-digit value for "works".

> oddball multi-pole cascaded filter sections

Right. I can show on the napkin that a filter will "work", and that F varies with these parts and Q varies with those parts. Or these-and-those parts, all interact-y.

But to compute the exact response, with component variations, requires drinking too many martinis (to get enough coktail napkins). That's where a stupid computer helps.

Another filter problem: empirically-designed tone controls. The James and Baxandalls are easy to figure, but the Fender tonestack is whacked. Can't call it stupid: been in continuous production for discerning users for almost 50 years. I can approximate it on napkin (and Kuehnel has a rigorous matrix-solution {which would be too labor-intensive before computers}), but drop it in SPICE and you can see the actual shape.

> subtle component quirks like with the good old cheap 3086

We are at the mercy of the models. Which, even more than data-sheets, put the best face on an imperfect part, and omit the naughty-bits.

The 3086 could be awful funky.

And then there are pentode vacuum-tube models. Often the screen current has nothing to do with reality, except in a very conventional circuit not driven to clipping. Even then, I was startled last week when a real 6550 sucked an Ig2 only 2:1 off from a model I had. Normally it hardly matters; I was dorking with an absurd loading and a goofy G2 supply where G2 current mattered (but in good cocktail-napkin thinking, not enough to make it stop working).

However, SPICE encouraged me to optimize for the model. When I beat on an actual tube, I decided I liked a different optimization. And when I then tried 5 different tubes from different factories and decades, 4 worked the same and the 5th worked 10% better, and a different-type tube worked 5% worse. Optimizing for the model gave a very fussy condition (1.000 versus 1.001 indeed) while working with actual tubes quickly led to a condition which all actual tubes liked. (Well, not SO quickly... I kept going back to what SPICE liked, instead of observing reality.)

I must say: when I measured power transformer regulation on the prototype, and put it back in SPICE, power output and approximate THD were right-on, near-enuff. And when I tried cathode resistors from 100 to 470, and loads from 5K to 20K, it tracked the saggy breadboard. If you know all the answers, and can stick them in SPICE, it sometimes agrees with reality.

> these phone calls saying "your part oscillates!"

My SPICE sim didn't oscillate. My prototype did. The nice thing about the prototype: by sticking a graphite pencil in it, I could get totally bizzare changes of squeal. That quickly narrowed-down the area of interest to two wires maybe too close, I shoved one, problem cured. (OK, I shut-down and re-dressed the wires 3 times further apart than the barely-fixed distance, then moved a ground wire between them.)

> "your part oscillates!"... the person hadn't actually built anything yet...

The person should not even be playing with electronics, if the person can't tell reality from fantasy.

Which is a problem for the younger generation of EEs. They don't play with real devices nearly enough, even for Graduate Thesis. I had a couple Honors EE kids in the electronic music lab, they needed one more fixed low-pass. I tossed them a box of caps and told them the synth had 1K output impedance everywhere. They didn't understand. When I explained, they said they didn't think it would work. They'd developed and demonstrated a bucket-brigade pitch-shifter (when B-B devices were just sampling), with the active filters such contraptions need, but couldn't picture a simple passive low-pass.

Oh: their pitch-shifter worked only a few seconds an hour. Their breadboard was MUCH too sloppy. They'd actually gained a lot of skill at "find the short". They could find 2 shorts per minute. Unfortunately the white hole-board was SUCH a mess, that it tended to accumulate 2.2 shorts per minute.

Back In The Old Days (uphill both ways through the snow winter and summer), we couldn't be so sloppy. When you do a cat-whisker radio, the cat doesn't give you a second chance.

 

[PRR]

> I second the use of a proper comparator (like 339). Using an opamp as a comparator may require dealing with input bias currents.

But Wayne's plan uses the 339s as voltage-followers. And the 339 has bias current (not enuff to hurt here). And he uses the no-current TL074 as the comparators.

I have no doubt it works. 339 is not stable as an op-amp, the fix is to drop a honking big cap on the output, and he's got a tantalum. And the 339 is rated to stand-off the large voltage between inputs. You ain't supposed to do that to op-amps, comps expect it.

It is possible to do that job without the 074, but the adjusting becomes uglee. Here he could even adjust both channels with the one pot.

I would think that, if you really want "clip warning", the threshold should bob with the rail, not fixed. The 24V BBC stuff was expected to run on battery in emergencies. Of course if your batts are sagging from 28V toward 19V, a little clipping may be the least of your problems.....

> tendency for a little oscillation burst at the trip point

Yeah, been bit by that. But a clip light, in this sense, is not a precision part, ideally will never trip, and IAC who will notice a 1MHz flicker in the LED?

Of course if that leaks out and beats against the tape bias oscillator, it may add a puzzling "color" on a hard-pushed track...

 

[Swedish Chef]

[quote author="PRR"]> [...I was startled last week when a real 6550 sucked an Ig2 only 2:1 off from a model I had. Normally it hardly matters; I was dorking with an absurd loading and a goofy G2 supply where G2 current mattered (but in good cocktail-napkin thinking, not enough to make it stop working). [/quote]

Do tell! :razz:

chef

 

[jdbakker]

[On a diode-less peak detector]

[quote author="mediatechnology"]See: Here

When I was doing meters I pretty much stopped using diode-based detectors 'cause this topology way out-perfomed it. With AC-coupled inputs to both comparators the low-level performance is pretty decent too though not needed in a clip indicator.[/quote]
Very nice design, Wayne. It reminded me of something I'd seen before, but it took a few days for the penny to drop.

This article (click past the annoying ad, or wait 20 seconds) uses a similar approach with an ECL comparator to detect 5ns peaks.

JD 'just another datapoint' B.