Discrete vs. Non-Discrete

[pucho812]

the simple answer is yes. But if you look at the hybrid opamps where it is an IC chip followed by a discrete transistor drive stage, you can make it happen.

 

[ruffrecords]

but is it correct to say that those capi preamps cannot be used with a 5534 in it because it cannot drive the output trafo and such?

Yes and no. The 5534 will drive a 600 ohm load so you can use a 1:1 transformer for a balanced output.  Neve and API amps use a step up transformer so they increase their maximum output level. This means for a 1:2 step up they need to be able to drive a 150 ohm load which the 5534 cannot do. So with +-15V rails you will get +22dBu into a 600 ohm load with  5534. With a Neve or API you can get +26dBU. If you have a discrete op amp that will take +-24V rails then it could drive also +26dBu via a 1:1 transformer.

Some discrete op amps can drive very low loads. I think the Neve BA440 can drive a load as low as 4 ohms. It was sometimes used to drive a cue speaker in some broadcast console designs. I think it was capable of about 4 watts.

Cheers

Ian

 

[abbey road d enfer]

With the risk of being flamed, I always thought you really cant do much better than a NE5534/NE5532 considering its price,

I would mostly agree with your statement, however, there are some circumstances where a 553x is not the most adequate choice. Among the reasons for not favouring a 553x are:
When the load impedance is excessive; that's the case when driving a 1:2 step-up xfmr with 600r load at the secondary.
When the opamp is used in a node that's extremely sensitive to input bias current; that's the case in the side-chain of many compressors, where a BiFET type will be preferred.
When the source impedance is way off the OSI of a 553x ( typically about 10-14kohm); that's the case with summing amps, where the bus impedance is typically under 1 kohm (300 ohms for 32 channels). A VLN opamp (ca. 1nV/sqrtHz), a combo of opamp and  BJT running at 1mA quiescent or a fully discrete design are to be preferred. That's also the case when the expected source impedance is very high, such as DI input; there a JFET input device will be preferred.

  the amount of API discrete opamps I've seen die is simply staggering, even more often than vacum tubes, most of the time I've seen clients replacing them with 990s just to get things going.

They do that without considering the fact that the 990 has an OSI of 1.1kohm, which does not make it particularly suitable for the typical 1:10 xfmr mic pre or even for EQ. However, it will improve the noise performance a in a summing amp.

So, considering today's integrated opamp technology, remind me, why do we need discrete opamps ?

ATM, only them (or similar hybrids) can offer 1nV/sqrtHz and 75r drive capability. For what it's worth...

 

[JohnRoberts]

ATM, only them (or similar hybrids) can offer 1nV/sqrtHz and 75r drive capability. For what it's worth...

OPA1622 specs driving 32 ohm  (2.8nV rt Hz ein).
OPA 161x  specs 1.1nV rtHz ein  and +/- 30 mA drive capability

The uber low noise op amps require the current capability to drive lower resistance feedback resistors to not give up the low ein benefit, however this is a bit of a paper chase since output noise will generally be dominated by source.

JR

 

[clintrubber]

there are some circumstances where a 553x is not the most adequate choice. Among the reasons for not favouring a 553x are:
When the load impedance is excessive;

Let's for completeness also mention The 5532 OpAmplifier:
a Douglas Self 5532-based poweramp: 32 times a 5532 in parallel, summed by 1 Ohm resistors, for ~15W in 8 Ohm...

... I do agree though this is mostly an amusing exercise, and not necessarily the most practical route.

Google for (pt1):  "Elektor-f8e460fa-81d3-410c-b7b3-8ce6f346dab8.pdf"

 

[Rob Flinn]

There can't be many recordings that haven't be through a 553x chip since it became commercially available.........

 

[JohnRoberts]

There can't be many recordings that haven't be through a 553x chip since it became commercially available.........

+1....  ;D

I recall flaming audiophools complaining about the "sound" of switch contacts..... How many switches do think are in recording console audio paths.  :

JR

 

[JohnRoberts]

From what I've heard from endless testing and obsessing is this.  Certain IC opamps do have a certain speed and frequency response that some either like or dislike.  NE5534 hasn't been my thing so far.. and I've given it lots of time to win me over.  I just tested NE5532, which is supposed to be the same as the NE5534 (times two) and without the comp/bal.  I preferred NE5532 for whatever reason that is, I don't know.  It's still not as nice as the BA283 discrete output design to my ear, by at least 50% for me, but everyone likes something different.

I was trying to like an unknown Class B discrete opamp design and prototyped it once with the iron, and about 5 times on the breadboard.  Made tweaks to try and turn it into something I thought I'd enjoy listening to over long periods, and it just didn't pan out.  Someone else might love it though.  It's more like the speed/cleanliness of a IC opamp.  So is it good, yes it is.  Anything else comes down to my preference.. and cannot be claimed as better or worse.

This sonic stamp preference, whatever that may be, seems to come down to Total Harmonic Distortion, and speed ... even for capacitors.  I'm still learning and don't claim any of this to be what anyone else will like or have to agree with obviously.. ICs are great.. convenient, cheap, sound nice.. and it's true... they're everywhere.  SO, no one can argue and say they're not effective or aren't already on countless songs that sound amazing.

A discrete design can have a similar THD to a IC and then there's really not much to talk about.. so isn't the discussion really more about how the distortion overlaps and the rate at which it moves and affects the frequencies.. and whether or not that moves you?  That's my conclusion at the moment anyway.

Adam

Op amps can have a characteristic transfer function open loop (without negative feedback), but once you apply negative feedback, all non-ideal (imperfect) transfer function characteristics are reduced proportionately by the "loop gain margin" or ratio of open loop gain to closed loop gain.

So yes, op amps can have a characteristic sound, but this is typically reduced to insignificance (in prudent designs). The sound of most circuits using negative feedback will be dominated by that negative feedback network, including non-ideal behavior from passive components in the NF network.

Op amps generally exhibit declining open loop gain at high frequencies, so operating with too high closed loop gain can typically show up as phase shift in the top octave or higher, while I have seen orders of magnitude larger phase shift caused by non-deal capacitors in gain legs than from the op amp open loop transfer function. (I've yet to see a compelling demonstration of audibility of this phase shift, but less is always better, no matter how audible.)

By these comments I am not supporting the concept that large audible differences exist between properly executed designs, just that there is a sliver of truth behind all this.  As has already been shared the largest benefit of discretes is higher voltage/current/power than typical IC processes support. BTW we do not need that much voltage/power to process useful audio signals.

JR

 

[ruffrecords]

Op amps can have a characteristic transfer function open loop (without negative feedback), but once you apply negative feedback, all non-ideal (imperfect) transfer function characteristics are reduced proportionately by the "loop gain margin" or ratio of open loop gain to closed loop gain.

JR

I agree with everything you say; I just want to add one caveat. The vast majority of op amps use a dominant pole technique to ensure high frequency stability. However, in very many cases, the turnover frequency of this dominant  pole is a few Hz or maybe a few tens of Hz. This means the open loop gain at 20KHz is often 40dB less than it is at 200Hz. This means the loop gain margin also drops by the same amount over the same frequency range and so the benefits of NFB deteriorate by the same amount.

One effect of this is that distortion rises by 20dB per decade across much of the audio band.

On the other hand, quite a number of discrete op amps set their dominant pole turnover frequency a lot higher, and sometimes above the audio band. This means the benefits of NFB are spread equally a across the audio band. The downside of this approach is that unity gain stability is often not possible (Neve BA440 for example).

Cheers

Ian

 

[JohnRoberts]

I agree with everything you say; I just want to add one caveat. The vast majority of op amps use a dominant pole technique to ensure high frequency stability. However, in very many cases, the turnover frequency of this dominant  pole is a few Hz or maybe a few tens of Hz. This means the open loop gain at 20KHz is often 40dB less than it is at 200Hz. This means the loop gain margin also drops by the same amount over the same frequency range and so the benefits of NFB deteriorate by the same amount.

I believe I said that but in the context of phase shift.  The dominant pole compensation makes the open loop transfer function look like a simple integrator with 90' phase shift.

Yes different op amps look different open loop and the 553x uses a split pole with some midband funniness... BUT when used prudently all the open loop funniness is well managed.

Do not use TL07x for high closed loop gain... duh....  Note this close loop gain also affect performance of VE summing amps. In sum amps it's called noise gain but effectively same thing as closed loop gain for loop gain margin evaluation. 

One effect of this is that distortion rises by 20dB per decade across much of the audio band.

yes, but the point is to keep it small even at 20 kHz...  back in the 70's I modified my bench SMPTE IMD tester to use 19kHz and 20kHz  instead of 60Hz and 7kHz, specifically to stress op amps in top octave...  If you try to measure THD at 20kHz the distorion prodcuts are reduced by the falling gain, two tone IMD creates a IM product down inside the passband.

This is all old news I hope.

On the other hand, quite a number of discrete op amps set their dominant pole turnover frequency a lot higher, and sometimes above the audio band. This means the benefits of NFB are spread equally a across the audio band. The downside of this approach is that unity gain stability is often not possible (Neve BA440 for example).

Cheers

Ian

As long as the errors are vanishingly small I do not care where the pole is.

I have heard this argument before (it was pretty big in the 70s) and I am a "show me" kind of guy which is why I modified my IMD bench test to stress the circuitry at 20kHz.  It was very revealing for the state of technology back then. Now we have much better ICs but some of the old soldiers are still respectable when properly used.

JR

 

[ruffrecords]

I have heard this argument before (it was pretty big in the 70s) and I am a "show me" kind of guy which is why I modified my IMD bench test to stress the circuitry at 20kHz.  It was very revealing for the state of technology back then. Now we have much better ICs but some of the old soldiers are still respectable when properly used.

JR

I am happy to concede that in a good design this should not be a problem. The problem is many designers did/do not take this properly  into account. My point is real life op amps are not like the ones you are taught about in school/university. There are plenty of wrinkles that set proper engineering apart from classroom theory. Op amps can be a temptation for an average engineer to simply drop one in and expect it to perform.

Cheers

Ian

 

[abbey road d enfer]

OPA1622 specs driving 32 ohm  (2.8nV rt Hz ein).

It's only marginally better than 5534 in terms of noise and 10dB worse than the best.

OPA 161x  specs 1.1nV rtHz ein  and +/- 30 mA drive capability

Which is marginally better than a 5534 in terms of output current but cannot compete in terms of voltage cause only +/-5V rails.

As I wrote earlier: for what it's worth... There are enough cases where the noise performance of 5534 is inadequate but the +20dBu output is necessary.

The uber low noise op amps require the current capability to drive lower resistance feedback resistors to not give up the low ein benefit

It's a matter of gain; expecting 1nV/sqrtHz @ 20dB gain is a losing battle. But perfect for summing amps that operate at about 30-40 dB noise gain.

 

[ruffrecords]

I also tried using a small value resistor (100R and also 510R) following the IC output before the output transformer, which is 600:600. 

A

That will just reduce headroom.  You do not need to match the transformer rated impedance. Preferably drive it from a much lower source impedance , which the 5532/4 is.

Cheers

Ian

Cheers

Ian

 

[abbey road d enfer]

I also tried using a small value resistor (100R and also 510R) following the IC output before the output transformer, which is 600:600. 

That's a recipe for LF distortion and attenuation. For best LF response, you need very low drive impedance. Good designs use negative-impedance drive.

 

[ruffrecords]

Just observations from a fascinated individual.  I would like to know if it's possible to reduce the small amount of DC hum I think I'm hearing with the BA283 circuit, other than getting gapped transformers.  Anything that can be done without changing the characteristic sound?  I unbalanced the transformer output in this test so maybe that's why the hum is there... only hear when I crank the monitor..

I will look up more about Negative Impedance Drivers.  Didn't realize there was such a thing.

Adam

The BA283 output stage  has zero power supply rejection. Noise in the power supply appears directly in the output. That is one reason why the amps were usually decoupled with a 1000uF capacitor.

Cheers

Ian

 

[efinque]

I remember reading somewhere (sorry, this is very vague) that some op-amps have a problem when they heat up (and some of them have a problem *because* they heat up)

I used to look at NE5534 and OPA134; there are others like the TL and stuff but the two were considered the recommended op-amps for projects at the time (2007ish)

EDIT : They're not really expensive or anything, it's just that for a bigger project where one would need a lot of them it's better to consider whether it's worth it, although with big mixing consoles for example I'd imagine the noise adds up when there are a lot of channels.

EDIT 2 : I just looked up and the OPA134 is almost four times more expensive than the NE5534; not really an issue in a small preamp or a DJ mixer but if one was to make a 48 channel console it would definitely be a factor.

 

[abbey road d enfer]

I remember reading somewhere (sorry, this is very vague) that some op-amps have a problem when they heat up (and some of them have a problem *because* they heat up)

ALL electronic components have problems when they heat up, until the magic smoke that's in them makes them tick is released... 
More seriously, yes, there are some issues with temperature, that's why there is a recommended operating range in the specs (-40 to +125°C for OPA134). It should not be a serious issue in the context of an audio product used in domestic or studio environment.

  They're not really expensive or anything, it's just that for a bigger project where one would need a lot of them it's better to consider whether it's worth it, although with big mixing consoles for example I'd imagine the noise adds up when there are a lot of channels.

Noise adds up, indeed, but in proportions that are easily predictable. For example, a TL072 is about 4 times noisier than a 5532, so, if a mixer using 5532's throughout was converted to TL072's, it would be 4 times noisier (about 12dB more). Or not...because all the other noises, particularly hum interference and noise due to PSU would still be unchanged, or even lower because the current draw of TL0 is about half of the 5532.

I just looked up and the OPA134 is almost four times more expensive than the NE5534; not really an issue in a small preamp or a DJ mixer but if one was to make a 48 channel console it would definitely be a factor.

There is no significant advantage in replacing 5534 with OPA134 (or 5532 with OPA2134) in a console because the current draw is just about the same and their noise voltage is about 6dB more. The OPA134 family beats the 5532/34 when presented with large source impedances (>50kohm) or when their lower THD is essential.
The two cases where I used them was for DI, where it has a slight noise advantage over a $0.20 TL072, and in instrumentation, where the disortion figure must be (hopefully) better than the DUT.

 

[Rob Flinn]

If a 553x is used properly there is no issue with noise.    For example if you have ever used a Raindirk Symphony it is one of the quietest desks out there.  You can turn the monitor control right up with no signal and there is nothing.  It is entirely filled with 5532 chips.      It amuses me when people start comparing the noise of chips to a 553x

 

[efinque]

Sorry I remembered wrong that the 5532 would've been mono but it was 5531 which is why I said 5534.

EDIT : fixed

 

[abbey road d enfer]

Sorry I remembered wrong that the 5532 would've been mono but it was 5531 which is why I said 5534.

EDIT : fixed

There is nothing like a 5531. The original (single) was TDA1034 from Mullard/Philips/Signetics, which became 5534 when licenced to other foundries. The sibling is the dual 5532. There was also an extinct 5539 (single in 14-pin DIL package).