Constructive wanted criticism for op-amp circuit design.

[beaker353]

I’ve attached a PDF of a project that has been in the works off and on for the past few weeks.  As I am definitely playing outside my usual prevue, I would like constructive criticism from the fellow board members.  I took a dc class back in high school in 1998 and in the spring a more formal dc class at a local community college.  I wanted to take ac this semester toward my EET, but my work schedule wouldn’t play nice.  I understand the very basics of op-amps, the idea of feedback, and the concept behind working from a single pole power source.  I have the educational version of Multisim 11 and modeled the TL074/TL071 part of the design and it is stable and acts as I would like it.  I have built the LM384 side of the circuit on a breadboard and it seems to also work as I would like.   Where I am having a bit of difficulty is choosing cap values and type (ceramic, electo, film, polarized, non-polarized).  I know Multisim says it should work, but I have a feeling the values I chose are not optimal given I was just choosing a number at random.

- The basic concept of the circuit is to monitor four audio signals on a local small speaker.
- Power (+30VDC) comes paired with each audio signal from a shielded twisted pair cable.  Shield as ground, red wire is power, black wire is signal.
- The diodes are there to prevent power from one line backing up onto another.
- The audio signal is speech (only so say about 500Hz to 4kHz) and maxes out around 1Vpeak.
- From time to time a positive DC signal (+15VDC) will be present briefly (1-10 seconds) on the audio line that needs to be filtered out reasonably well from the op-amps and power amp.
- I must be careful how much load I place on the signal line as it is a bussed system that many many components share.  For that same reason, I need to be careful about adding anything (like DC for instance) back onto that bussed signal line.
- VR1 through VR4 act as volume controls for each individual channel. VR5 will be a trim pot to reduce the strength to of the signal going into the LM384 as it has no gain control.
- As power is from a big beefy central power supply, power consumption within reason isn’t a big deal and is very well filtered.

Let me know your thoughts, but please be nice…  Thanks!

- EM

[ejeffrey]

C4-C7 set the low frequency rolloff.  They also will be responsible for blocking the DC pulses, and the bigger they are the bigger the thump when the DC switches on or off.  Out of curiosity, why is there a switching DC level on an audio signal line?  Anyway, the lower cutoff frequency is 1/(2*pi*R1*C4) or 16 Hz.  Since you are only handling voice and probably don't have much in the way of bass response in any case, you can bump that up.  Try a smaller capacitor.  330 nF will give you around 50 Hz high-pass.

Your virtual earth could be a bit better filtered.  I would make the C15 10 uF and probably make R14 and R15 10 kohm.  That will reduce the coupling of any supply ripple onto the audio.

U3 has the inverting and non-inverting terminals switched.

U4 shows the non-inverting input connected to GND, I think that should be V/2 unless the LM384 is internally capacitively coupled (I have never used it).

Your volume control is not really optimal.  It gives a gain range of only 2:1.  For audio work you usually want more like a 10:1 ratio or more using a pseudo-log pot so you have useful sensitivity across the whole adjustment range.

C8-C11 and C13 are unnecessary.  No reason not to use a DC coupled signal path here.  As for the other capacitors, C4-C7 can be any type.  Hi-fi people generally prefer film caps for the audio signal path, but it doesn't really matter for AC coupling capacitors (the AC voltage across them is very small), and you aren't doing hi-fi in any case.  If you use electrolytics, make sure to put them the right way around so that when the DC voltage is present they are correctly biased.  Most of the rest of your capacitors are power filter/bypass caps and should be electrolytic for large values and ceramic for small values.  C18 is fine as shown, and gives you a 50 Hz high pass on the output.

[JuKu]

U3 needs both inputs connected. Put a small cap in parallel with R5-R8 and R13 to minimize hiss and increase stability.

The DC pulses will be a problem if the edges are sharp. If so, you'll have a full volume bang and no easy solution. Think it like a 15x full volume square wave (although only the first part of it). You do want a good (40-60dB) attenuation at that frequency. If the pulse raises and falls in 1sec, the project has a change. If the rise time is 1ms, you are in trouble.

[Gall]

There's something wrong with U3 - it has no negative feedback and a dangling input. Connect it like U2 (swap "+" and "-" inputs, ground the "+").

As mentioned above, C8-C11 and C13 aren't really necessary. You already have C4-C7.

I'm not sure LM384 doesn't need negative feedback, please check the datasheet.

You may improve performance by adding an active voltage follower at V/2 betveen the R14 R15 C15 joint and rest of the circuit. To do that, replace TL071 with TL072 and you'll get one more opamp. This is not very necessary but may reduce noise.

Another possible approach is to use 7812+7912 instead of 7824. So you may get rid of C18 and avoid click at power on. You also may use BTL output for the same purpose.

Replacing LM384 with TDA2030 or similar would simplify the output stage.

[vk6zgo]

I was going to say "You've come to the wrong place for constructive criticism" ,but all the posters are on their best behaviour!

VK6ZGO

[beaker353]

Going back a forth between testing in Multisim and then building in DipTrace I had a few typos.  I thought I caught them all before I originally posted.  Figures.  New posting coming in a few minutes...

- EM

[beaker353]

Thanks for all the input.  I knew that the request “constructive criticism” was dangerous, but sometimes you get exactly what you ask for.  Which is exactly what I got…
Let’s see...

ejeffrey:
-Now that formula for cutoff frequency makes more sense.    Even 250Hz would work for the human speaking voice, but 50Hz sounds like a good place to start.
- The DC is for an indicator light.  This is connecting bussed intercom system that is common in the entertainment field.  If someone isn’t listening to a channel for whatever reason pushing a button on your intercom pack will send the DC on the line and a bright light will light up on all the packs connected to that line to get their visual attention.  Not the most high tech, but keep in mind this concept was designed in the 1960’s and has changed little since.  Still runs many of the largest production shows in the world (and a particular one I work for in Las Vegas)
- Increasing C15 makes since.  As well as R14/15.
- U3 was a big typo as I was transferring my design work in back and forth between Multisim to DipTrace. 
- The LM384 is a little power amp chip that specially designed to run from a single sided supply without having to bias an input.  All the diagrams and application notes I have been able to find ground out the non-inverting input.
- As the gain on the LM384 is hard set at +34dB (one thing I dislike about it) and the signal coming in from the audio lines is pretty hot, I doubt that I will need any amplification of the signal going into the LM384.  More than likely I will need to attenuate the signal some.  VR5 should give me the ability to pad about 25dB or so as needed, I think.
- When I remove C8-C11 from the Multisim simulation everything seemed to play nice.  I now had various levels of DC present after the 10kohm mixing resistors, but C13 blocked those from getting to the TL071, no problem I guess. However, once I removed C13 and had VR1-VR4 at anything less than full tilt the final waveform was all messed up.  I’m guessing that the DC bias on U3 was causing some type of rail clipping.  I know Multisim isn’t the end-all and be-all, but it makes since.  Maybe someone can clarify what I’m seeing in the simulation.
- Then would I be better to make C14 a 330nF non-pol  as it is a DC blocker and not a filter cap?

JuKu:
- U3 was typo.  A big typo, but a typo.
- The caps across R5-8 & 13.  Are you suggesting 330nF’ish or more like 100nF’ish?
- Never actually measured the rise and fall times, but I image they are in the order of 250mS or so.  I’m going to do some testing with the scope tomorrow I guess.

Gall:
- U3, I know, I know…
- See my above response to ejeffrey about my cap confusion.
- Ditto about the LM384.
- I was looking at some application notes about that earlier.  I have some extra TL071 on the way.  Worth a try to see if it adding it improves the performance any.  Stay tuned...
- Interesting idea on the 7812 + 1824.  But I’m not sure if that would work with the LM384 as it requires a single sided supply and it needs to be in the 20’ish range for the rails.  Further research needed
- The TDA2030 looks a little over-beefy for my desktop application, but I can think of where that much power could be useful.  I’ll need to put a couple on order to play with.

Vk6zgo:
I’m a little surprised of everyone’s niceness as well.  Not that I don’t enjoy what I got back, because I REALLY do.  Thanks all! See attached "Rev B".

- EM

[Gall]

- Interesting idea on the 7812 + 1824.  But I’m not sure if that would work with the LM384 as it requires a single sided supply and it needs to be in the 20’ish range for the rails.
Further research needed

Yep. If you use 7812 and 7912, you'll still get 24V between the rails. You just replace "fake ground" (V/2) with "real ground" (and your present "ground" becomes "-12V" then). In most analog circuits with opamps its better to have bipolar power (+12 - GND - -12) than unipolar (+24 - GND). Sometimes it may get problematic to couple them with unipolar circuits but any opamp would do the trick. Since center GND is powerful now, it is possible connect the load to it directly without capacitors, provided that DC is properly cut off before the output stage. This design is very usual.

Sometimes you're limited to unipolar power, i.e. if your GND should be connected to a metal case (in car amplifiers etc.) but you still don't want big catpacitors at the output. This can be solved using bridge output circuitry. Depending on desired power, the output stage varies from TL072 to a couple of TDA7293s.

- The TDA2030 looks a little over-beefy for my desktop application, but I can think of where that much power could be useful.  I’ll need to put a couple on order to play with.

TDA2040 and TDA2050 are very similar, too. I used them many times. They are just powerful opamps and may be used for many purposes other than audio. Even to make a regulated DC power supply. I use them for little power as well, just without heatsinks (TO220 is small anyway).

For many low-power amplifiers I just made the output styage from TL071 with two transistors (npn+pnp) for an emitter follower. (Both bases to opamp output, collectors to power rails, emitters form circuit output, opamp feedback goes to emitters instead of opamp's own output). This works fine. For small power (and low quality) there's no even need for any bias between bases. There's also a nice high-power circuit using TL071's power pins for a current output with "real" output grounded, we made big 1kW concert amplifiers using this technique. Works for smaller powers, too.

[ejeffrey]

Definite improvement

I didn't notice this the first time, but VR1-4 are connected to GND, not V/2.  That is why you get DC voltages and needed C13.  I would probably connect those to V/2, which then would probably need to be buffered with an op-amp.  Alternately, if you keep it the way it is, you would need to make sure C13 is polarized with the positive terminal facing U3 (if it is a polarized cap).

The capacitors C8..C11 and C20 in parallel with the feedback resistors are too big.  They cause a low-pass response and only at 160 Hz.  Those capacitors should be more like 1 nF at most.

I still don't  like the volume control VR5.  It is better now because you don't have the fixed 10k that limited the dynamic range in the old schematic (Edit: I misread the original schematic) a psuedo-log pot will give you better tuning over the whole range.  There are a number of ways to do this, for instance see figure one in the link below.

http://sound.westhost.com/project01.htm

The idea is that each degree of rotation provides approximate the same number of dB attenuation, which is approximately what you want for an audio volume control.  It gets you a wide tuning range without being too sensitive at one end and too insensitive at the other.  You have actually already done this on VR1-4 (once you realize that U3 is providing a virtual earth, the circuit is exactly like the esp page I linked to).

[Gall]

The capacitors C8..C11 and C20 in parallel with the feedback resistors are too big.

They aren't in parallel with feedback resistors.

[baljemmett]

I’m a little surprised of everyone’s niceness as well.  Not that I don’t enjoy what I got back, because I REALLY do.

Generally if someone is asking sensible questions and demonstrating willingness to listen to and understand the advice they receive in return, knowledge flows freely.  It's when one or more of those prerequisites appears to be lacking that people end up getting frustrated and the arguments start   The same thing can be observed in many other fields and forums too, of course, so it's not unique to this place!

[ejeffrey]

The capacitors C8..C11 and C20 in parallel with the feedback resistors are too big.

They aren't in parallel with feedback resistors.

It wouldn't be the first thing I have misread in this schematic, but I am pretty sure they are (in rev b).  100 nF capacitors in parallel with the 10k feedback resistors (R5-R8) of U2.1-4, and also C20 in parallel with R13.  These will set a low-pass filter for the amplifier that will heavily filter the desired signal.

[ejeffrey]

I was going to say "You've come to the wrong place for constructive criticism" ,but all the posters are on their best behaviour!

VK6ZGO

Are you kidding?   We should frame this post to show new people looking for circuit advice.  He clearly laid out his goals, provided a schematic of his current state of progress, and his expectation of how it would work, and asked specific questions about things he was unsure of.

[Gall]

The capacitors C8..C11 and C20 in parallel with the feedback resistors are too big.

They aren't in parallel with feedback resistors.

It wouldn't be the first thing I have misread in this schematic, but I am pretty sure they are (in rev b).  100 nF capacitors in parallel with the 10k feedback resistors (R5-R8) of U2.1-4, and also C20 in parallel with R13.

Oh yes, I looked at wrong schematic.

If using TL07x opamps, these capacitors are not really necessary since TL07x work fine and stable without any such correction. They may be needed if using different opamps.

[blackdog]

Hi beaker353

This is my 2 cents,
I draw a schematic for you, is uses 1 opamp less...
www.bramcam.nl/ClearCom-V2.JPG

I used a little higher value for the resistors because the 15V DC pulse to protect the inputs.
C1, C2 and C12 wil give you a highpass filter about 400Hz.
R9, 17, 20, 23 and P5 + C9 make a lowpass filter around 6Khz.
P1, P2, P3 and P4 are Preset, P5 is the Master volume control.

This is the best i can do and keep it simple, 15V Puls on audio line Brrrrrrrrrr :-)

Kind regarts,
Bram

[ciccio]

I build intercom stations for a living, and I've some experience with them (I've built my first one in 1983).
I ask the permission for some suggestion, in this maybe too long post..

first: ejeffrey is correct: C8, C9, C10 and C11 are in parallel with the feedback resistors of the input op-amps.
They should have a value in the range of 100 pF, not 100 nF.

second: I'll suggest  4  polarized caps (About 10 uF)  at the outputs of U2.1 A-B-C-D.
This will prevent DC flowing on the pots, resulting in noise when the pots are rotated.
C13 should be a polarized cap in the same range (10 uF). The positive pin of the caps must look to U3.
Leave the pots wired to ground, not V/2, because  V/2 is not a "zero impedance ground", and this will cause crosstalk and interaction of the 4 pots.
To reduce noise when rotating the master gain pot, a capacitor (100 to 330  nF) should be wired is series with VR5 wiper, connecting to LM384 pin 2, and this pin should  be grounded via a 10 - 100 k resistor (this is not shown in the application sheet of the IC, but is a good practice).

Blackdog's modified schematic can work, but the passive mixing of the four signal will result in interaction of the controls (changing one channel's gain will change all the others).
Another effect is a reduction of total gain, due to the loading of the pots by the summing resistors.
Another op-amp will work better, and will allow for total gain adjustment by changing it's feedback resistor.

If the power is supplied by a ClearCom power supply (30V DC),  you do not need a voltage regulator (but it will do no harm): a simple RC filter (10 ohms + 47uF) at the DC input will suffice (the series resistor will act as a fuse in case of failure, putting your station "out-of-service" and allowing normal operation of the rest of the system). Your circuit has no such protection (even a small fuse will work).

In my experience, the LM384N power amplifier will work at 24 or even at 30 Vdc, but it's a good idea to put a 10 to 22 ohm, 2W resistor from power supply to pin 14, and a 100 uF cap in parallel with C16 (see any ClearCom schematic).  This will limit output power, power dissipation and current drawn from the intercom line.
C18 value can be reduced to 100 - 220 uF. This will limit LF response and power dissipation in the LM384.

Take good care of the  PCB: all ground pins of LM384 must be soldered to a small ground plane, to help heath dissipation. If you want to build on perfo-board, glue a small heatsink  to the IC.

I believe that, instead of one quad and one single op-amp, 3 dual (TL072 or even better NE5532) will allow for a simpler layout. The spare op-amp can be used for buffering the virtual ground (connect it's non-inverting input to the  junction of R14 and R15,  wire for unity gain by  shorting the inverting input with the output  and connect V/2 to the output).

The Virtual ground reference (R14, R15, C15  and maybe the buffering op-amp) will work, but it is a good idea to wire a diode in parallel to R15 (cathode facing positive supply): this will help to discharge C15 at power-off (some op-amps are damaged in this circumstance).

The LM384 is a better choice than the TDA2040 because it's current drain is only 8-9 mA , compared to the 100 mA of the latter, and because very few external components are required.

The "call signal" on audio line is derived by injecting a DC current on the line, which  has an impedance of 5 kohms at DC  and 220 ohms for audio frequencies.  The resulting voltage is usually no more than 5 V DC, and it gives an audible "thump", but it cannot damage any component, not even the speaker.
Usually the users wear headsets, and they are not annoyed by this.

It's one of the few possible solutions for  ClearCom compatible, "party line" intercoms, where the stations are connected using a standard microphone cable (2 conductors + screen) with XLR 3 pin connectors (male + female).
The screen (pin 1)  is ground, the conductors carry DC power for the stations (pin 2) and audio line (pin 3).
The simplest path for sending a call signal  is DC over the audio line (there no no free wires).
Some systems use ultrasonic signals to transmit other signals or commands , and other interrupt the DC power for a very short time to switch off all stations' microphones.
It is old analog technology, but it works, day after day, show after show...


Good work , indeed

Regards

[blackdog]

Hi,

Cicio,
The passive mixing is no problem at all, this is not a mixingdesk at a recording studio...
The interaction is less than 3dB and these are preset controls.
Yes, you lose some gain, also no problemo, the peak level of the audio is about 1V pp or say 300mV rms.
This is my version 2, now it's a LM386N4 as poweramp configured with extra gain and lower quiesencent current.
But also les power, about 1 Watt.
There is now some extra filtering at the input to keep RF out of the circuit.

C21, 22, 23, 24 Ceramics
C15, 16, 12, 14, 20 Electrolitics
Rest = Film

Keep the tracks/wires at the negatif inputs of the opamps Short.

www.bramcam.nl/ClearCom-V3.JPG


Shoot at it Ciccio ;-)

Kind regarts,
Bram

[ciccio]

Hi Blackdog,
your circuit should work correctly, when accepting the limits of passive mixing.
The 7812 will need some heat sinking.

Best regards

[Gall]

Even 3 capacitors less, a single capacitor after potentiometers would work fine.

However a buffer opamp between the mixer and volume control pot is useful. It would prevent influence of pot positions on each other.

[ciccio]

Even 3 capacitors less, a single capacitor after potentiometers would work fine.

No, the op-amps outputs are at 12 V respect to GND.
Without the caps DC will flow through the pots, and after some time will result in noise or "scratches" with pot rotation.
It should be possible to refer the pots CCW pin to the virtual ground (12 V DC), so there is no DC over the pots, but in my experience a small DC voltage is always present at the outputs of the op-amps, so it is better to have the caps.
The more the caps (by selecting slightly different cut-off frequencies at each filter), the better will be the low frequency cut-off.

However a buffer opamp between the mixer and volume control pot is useful. It would prevent influence of pot positions on each other.

That's true, see my previous posts.
With the schematic suggested by Blackdog another capacitor at the input of the LM384 can be a good idea: the input is not at ground potential, so the master volume pot could have the same noise problems.
best regards

[Zero999]

Blackdog's modified schematic can work, but the passive mixing of the four signal will result in interaction of the controls (changing one channel's gain will change all the others).
Another effect is a reduction of total gain, due to the loading of the pots by the summing resistors.

That can be minimised by more careful selection of the component vales.

Change R9 to R2 to 130k, P5 to 100k and C9 to 1nF.

[Gall]

Blackdog's modified schematic can work, but the passive mixing of the four signal will result in interaction of the controls (changing one channel's gain will change all the others).
Another effect is a reduction of total gain, due to the loading of the pots by the summing resistors.

That can be minimised by more careful selection of the component vales.

Change R9 to R2 to 130k, P5 to 100k and C9 to 1nF.

Yes, but not always. Also I don't think it's a good practice to rely on potentiometer values. They may have 20% tolerance or worse. An extra opamp costs nearly nothing.

Usually I take TL074s for such circuits and use extra opamps for VCC/2 buffering, active ground feedback or even in power supply feedback. Or I make everything on TL072s to simplify board layout if I need lots of passive components. I use TL071 only if I need deparate power pins (i.e. to play with supply current). I never use balance pins of TL071; if I need more precision, I always take more expensive opamp.

[blackdog]

Hi all 

The component value's are careful selected to be shure the "click" from the 15V pulse is damped as good as possible.
Yes i know it can be better, but at what cost...

The passive mixer is no problem, interaction is less than 3dB for "preset" potmeters!!!
I say it again, this is not a mixer in a recording studio.
I selected the gain so that you can just reach full power when all the pots are full open.
If there is more gain necessary, remove R27 and connect C16 to pin-8, this will give about 6 dB extra gain.
Still not enough, double the value of R1, 15, 18, 22.

Making P5 100K will give you 3,5dB extra gain, but you have to buy a extra value potmeter, and, do you need this extra gain?
A higher value of the pots will give more noise.
There maybe 20% tolerance in the potmeters value, this is not important, only a little shift in the corner of that filter.

Filtering
C1 +R2 = 220Hz
C2 + P1 = 160Hz
LS + C12 = 90Hz
This will give a nice rolloff below 400Hz

R9, 17, 20, 23, P5 and C9 = 6400Hz

R29 and C21 ~160KHz to keep the HF out.



Tell me, what is wrong 

Kind regarts,
Bram