A summing amp with more transformers

[bjosephs]

After being a little underwhelmed with the impact of my last summing amp project I decided to go back to the drawing board and make something with more influence on the sound. I looked into doing a passive network into a transformer coupled make up stage but the reality is that signal attenuation in the passive network prevents the iron in the middle from being stressed at all. So, I'm trying an experiment with a transformer at each and every input. This is costly, so I went cheap and imported 12 OEP Z21807C to try things out. I'm posting a first draft with things kept rather simple:

• The OEP connected as a step down "low" impedance and the simplest termination network
• An inverting 5534 make up stage with a gain of a little more than 3 to bring things back up
• A simple LCR pan network into L/R bus using a DP3T switch
• Virtual earth summing amps (5534 again, will need a comp cap on this stage) with a class a buffer borrowed from hifisonix which has a lot in common with the Neve Portico output stage found here

A couple things I kind of want to try but only but I'm talking myself out of:

• A fader/pan pot - but I don't want the complexity of pan buffers so I'd need the first 5534 to drive a 1k/dual10k combination... technically above its 600ohm capability but I'm not sure its a good idea
• A non-VE makeup stage a la the BA283 instead of the virtual earth stage but I'm not sure what this would do to my pan circuit
• The Neve texture circuit found in the link above but I'm not sure they aren't just trying to destroy transformers

Thoughts?
Brian

 

[ruffrecords]

The Z21807C is a great little transformer. I am using them for the tape return inputs on the latest version of channel amp on the Mark 3 Tube Mixer. If you are looking for serious colour then you may be disappointed. They will take a lot of level before noticeable distortion.

Cheers

Ian

 

[abbey road d enfer]

After being a little underwhelmed with the impact of my last summing amp project I decided to go back to the drawing board and make something with more influence on the sound.

I always find this approach a little suspect. My understanding is that you expect the summer to add to your sources something that was lacking. Shouldn't you start by processing the tracks individually? Applying the same paint/varnish/spices over all tracks seems inadequate to me.

• The OEP connected as a step down "low" impedance and the simplest termination network

I suppose you mean wiring the secondaries in parallels, for 6dB attenuation. Apart from a possible (?) extension of the HF response, I don't see any advantage.

• An inverting 5534 make up stage with a gain of a little more than 3 to bring things back up

An inverting stage is not the quietest solution, particularly if you want to go by the recommended load (100k). Why the choice of this topology? Don't tell me transient response, not in the same sentence as transformer.
EDIT: I see you have set for a 2.7k input, which is gonna ruin the HF response of the xfmr.
EDIT: I just saw the low-Z connection option.

A couple things I kind of want to try but only but I'm talking myself out of:

• A fader/pan pot - but I don't want the complexity of pan buffers so I'd need the first 5534 to drive a 1k/dual10k combination... technically above its 600ohm capability but I'm not sure its a good idea

You could use a 22k LIN pot. The law-steering effect of the load presented by the pan-pot would provide a good taper (-10dB at 75% rotation/travel) and load always greater than 2.5kohms.

• A passive makeup stage a la the BA283 instead of the virtual earth stage but I'm not sure what this would do to my pan circuit

"passive makeup stage" is definitely an oxymoron. Call it voltage summing amp, for correctness.
Regarding possible effect on pan circuitry, it's not more significant than with VE summing. However, there are consequences on X-talk.

 

[bjosephs]

The Z21807C is a great little transformer. I am using them for the tape return inputs on the latest version of channel amp on the Mark 3 Tube Mixer. If you are looking for serious colour then you may be disappointed. They will take a lot of level before noticeable distortion.

But they're so small!

To be clear I'm not looking to mangle things so maybe it will be the right balance. Care to share how you plan to implement them? 1:1 or 1:.5? Are you using the terminations from the datasheet?

 

[bjosephs]

I suppose you mean wiring the secondaries in parallels, for 6dB attenuation. Apart from a possible (?) extension of the HF response, I don't see any advantage.

I chose this for 2 reasons: It's the simplest termination (datasheet shows no need for a damping network) so I thought this was wise; my original plan was to build some BA283 style discrete gain stages which expect a -8db nominal input signal. Clearly I put that aside so it is not relevant. If using the 1:1 termination is likely to have benefits I will certainly go to that. If it's just the risk of a little more noise I'm not too worried.

An inverting stage is not the quietest solution, particularly if you want to go by the recommended load (100k). Why the choice of this topology? Don't tell me transient response, not in the same sentence as transformer.
EDIT: I see you have set for a 2.7k input, which is gonna ruin the HF response of the xfmr.

How can you tell that the 2.7k input will ruin response? I attached the datasheet for the transformer in my original post and while its not the "optimum" termination it isn't advised against. I needed a gain stage/buffer and since the bus amps invert I defaulted to inverting to maintain absolute phase. I can always invert with the transformers as well. What is the noise difference between inverting and non-inverting? Part of the reason I went with the lower resistor values was for noise reasons and to help control DC offset without coupling caps. I could change to a non-inverting with coupling cap, 100k to ground and wired for unity gain. Better tradeoffs?

"passive makeup stage" is definitely an oxymoron. Call it voltage summing amp, for correctness.
Regarding possible effect on pan circuitry, it's not more significant than with VE summing. However, there are consequences on X-talk.

Ya I don't know where my head was at there... mixing "passive" bus with "active" summing or what not. But as you say: voltage gain recovery vs. VE summing. Thanks.

 

[abbey road d enfer]

How can you tell that the 2.7k input will ruin response? I attached the datasheet for the transformer in my original post and while its not the "optimum" termination it isn't advised against.

You may see that I have edited my message after having read the 2nd page of the document, which I had neglected.
I think such a drastic difference between the two wiring possiblities is odd.

What is the noise difference between inverting and non-inverting?

In inverting mode, when loading considerations impose a high input resistance, the effects of input noise current become significant. If you wanted to use an inverting stage with the 1:1 connection and its recommended load of 100k, you would have 100k at the input and 200k in the NFB loop, that would result in 66k seen by the opamp's input. It would result in 3.7uV noise, to compare with the 0.5uV due to the input noise voltage (with a 5534).

Part of the reason I went with the lower resistor values was for noise reasons and to help control DC offset without coupling caps. I could change to a non-inverting with coupling cap, 100k to ground and wired for unity gain.

In non inverting mode, with reasonably low values in the NFB network and a low-Z source, you can expect noise due to the opamp to be less than 1uV.
But since it appears like the xfmr can accomodate the lower Z, why not?

 

[bjosephs]

You may see that I have edited my message after having read the 2nd page of the document, which I had neglected.
I think such a drastic difference between the two wiring possiblities is odd.

OK, disregard . I have asked OEP for any data they can share (Some of their datasheets have more detail than others) but I may just jump tracks and use the "optimum" configuration. Maybe Ian will chime in with his strategy as well.

In inverting mode, when loading considerations impose a high input resistance, the effects of input noise current become significant. If you wanted to use an inverting stage with the 1:1 connection and its recommended load of 100k, you would have 100k at the input and 200k in the NFB loop, that would result in 66k seen by the opamp's input. It would result in 3.7uV noise, to compare with the 0.5uV due to the input noise voltage (with a 5534).

I see... inverting mode risks high(er) noise because load resistance is in series with source resistance. This makes sense.

 

[MisterCMRR]

The Z21807C is a great little transformer. I am using them for the tape return inputs on the latest version of channel amp on the Mark 3 Tube Mixer. If you are looking for serious colour then you may be disappointed. They will take a lot of level before noticeable distortion.

Cheers

Ian

Conspicuously absent from the data sheet is the source impedance used for the specs (max level, THD, etc.). This is a critically important test condition. Any POS transformer will look worlds better if driven from a zero-impedance source. This is a very common omission from data sheets trying to hide the ugly truth ... and a pet peeve of mine!

 

[abbey road d enfer]

Conspicuously absent from the data sheet is the source impedance used for the specs (max level, THD, etc.). This is a critically important test condition. Any POS transformer will look worlds better if driven from a zero-impedance source. This is a very common omission from data sheets trying to hide the ugly truth ... and a pet peeve of mine!

I fully agree, with no second thoughts.
However, the OP wants xfmrs in the signal path in order to impair whatever they do to the signal. One's ear delight can be one other's sonic crap. Endless subject since it involves particular hearing experience and acquired tastes.
Similarly, an other thread praises the vitues of tantalum-induced LF distortion for the sake of conforming to historic sonic signature...

 

[MisterCMRR]

Absolutely agreed - there are two camps in audio: accuracy and "artistry" ... and likely they'll never agree, nor do they need to. But I have strong feelings about data sheets (having created all the data sheets at Jensen). There should be a clear line between accurate information and "marketing," where half-truth is just one tool used to sell it.

 

[JohnRoberts]

I am not the transformer expert here but I recall one casual discussion with Deane Jensen (RIP) several decades ago about distortion in transformers. He talked about metallurgy, and other stuff that was above my pay grade. As I recall he was adamant about driving transformers from low impedance.

This suggests to me (still not a transformer guy) that there is more than one source of non-ideal transfer functions in transformers, so trying to stimulate it(them) on purpose may involve multiple factors.

JR

PS: only two camps?

 

[bjosephs]

Conspicuously absent from the data sheet is the source impedance used for the specs (max level, THD, etc.). This is a critically important test condition. Any POS transformer will look worlds better if driven from a zero-impedance source. This is a very common omission from data sheets trying to hide the ugly truth ... and a pet peeve of mine!

Very common indeed, so I simply assume that achieving the spec requires approaching 0 ohms. The DAC that will be driving the inputs is spec'ed at 100 ohms. We'll see what that gets me. This is all for fun.

 

[abbey road d enfer]

Very common indeed, so I simply assume that achieving the spec requires approaching 0 ohms.

Actually, the right amount of negative impedance gives spectacular results for ouputs.

For inputs, I've always favoured current mode.

 

[MisterCMRR]

Note that, if you draw a simulation schematic for a transformer (in its simplest form) the primary consists of a resistor (DC winding resistance) in series with an ideal (distortion-less) inductor. All magnetic non-linearities (hysteresis and saturation) are modelled in a non-linear resistance in parallel with the ideal inductor. The total series resistance (winding + source) is in parallel with the non-linear resistance, so lowering it reduces non-linearity. But there's little to be gained by making the external source impedance less than 10% of the transformer's own primary DCR. However, this DCR can be cancelled (effectively made zero) by driving from a negative source impedance. This idea was patented by Audio Precision and used to produce the ultra-low-distortion output transformer in the Ap analyzers. To make such negative-output-impedance circuits stable requires compensating for the positive temperature coefficient of the transformer's primary DCR, so the transformer had a tiny-gauge zero-inductance winding for the compensation. Jensen supplied these transformers to Ap for many years. The patent expired long ago BTW.

 

[ruffrecords]

Conspicuously absent from the data sheet is the source impedance used for the specs (max level, THD, etc.). This is a critically important test condition. Any POS transformer will look worlds better if driven from a zero-impedance source. This is a very common omission from data sheets trying to hide the ugly truth ... and a pet peeve of mine!

I agree 100% which is why I tested them my self from a known source impedance. Lack of meaningful and detailed specs from transformer manufacturers is a pet peeve of mine too. However, OEP do seem open to constructive criticism. They typically quote primary inductance at 1KHz when what you really want to know is what it is at 100Hz or less. Several times I have requested this and they have measured it and updated their data sheets accordingly.

Cheers

Ian

 

[abbey road d enfer]

However, OEP do seem open to constructive criticism. They typically quote primary inductance at 1KHz when what you really want to know is what it is at 100Hz or less. Several times I have requested this and they have measured it and updated their data sheets accordingly.

My past experience with OEP left me with a rather poor opinion of their technical expertise. I must say, when I saw some of their newer datasheets, they have made an honorable effort.
Inductance at 100Hz is parcellar, since it varies significantly with level; however, it's more useful than the inductance at 1k, which has almost zero significance.

 

[bjosephs]

Well I finally got around to doing some breadboarding and attached are photos of the square wave response with no loading, series connected with 10k and no damping network, and parallel connected with 2.7k (no network required).

 

[bjosephs]

After a little more breadboard time I think I'm staying with the parallel connected secondary and will be using the NE5534 to raise the signal up. The NE5534 can run uncompensated at that gain so while there will be a bit more noise (compared to 1:1 connection) there will also be fewer parts to stuff. I've upped my mix resistors to 22k (I guess even 47k isn't uncommon) to lighten the load on the 5534 so I can use a resistor to bias class A... been wanting to try that.

I also tried a couple of output buffers using BD139 - the Hifisonix linked above and a different one biased with a ring-of-two (I believe its called). I prefer the elegance of the latter. The BD139 definitely need a heatsink to get full output into 600ohms with +/-15vDC.

 

[Matt Syson]

Hi folks
Interesting discussion here.
One of my (many) pet peeves is the notion that bunging any old transformer into an audio path WILL make the audio 'sound' better, whatever 'better ' might be. I find your talk of driving impedance interesting and useful but would like to add my thought/conviction that it is the characteristics of the driving amplifier, output impedance, whether it has symmetrical rise times AND how it reacts when faced with transformer 'reactance' when it heads towards the DCR at low frequencies. Does it current limit, like a typical op amp output would do when faced with a current overload where the internal protection circuits mostly take chunks out of the signal waveform in an attempt to protect itself so avoiding the chip manufacturers (embarassing) risk of failures. then with the possibility of the transformer inductance and capacitances how that modifies ant 'squared edges' that might be generated by such current limiting, that might actually be assymmetrical. thus transformer A, driven by amplifier design B, working into total load C representing a lot of variables.
Matt S

 

[abbey road d enfer]

Hi folks
Interesting discussion here.
One of my (many) pet peeves is the notion that bunging any old transformer into an audio path WILL make the audio 'sound' better, whatever 'better ' might be. I find your talk of driving impedance interesting and useful but would like to add my thought/conviction that it is the characteristics of the driving amplifier, output impedance, whether it has symmetrical rise times AND how it reacts when faced with transformer 'reactance' when it heads towards the DCR at low frequencies. Does it current limit, like a typical op amp output would do when faced with a current overload where the internal protection circuits mostly take chunks out of the signal waveform in an attempt to protect itself so avoiding the chip manufacturers (embarassing) risk of failures. then with the possibility of the transformer inductance and capacitances how that modifies ant 'squared edges' that might be generated by such current limiting, that might actually be assymmetrical. thus transformer A, driven by amplifier design B, working into total load C representing a lot of variables.
Matt S

Indeed, the improvements imparted by negative impedance drive is valid only whilst the circuitry operates in its linear domain. I have used this in thousands of products with just a properly implemented NE5532 and the performance is good enough to validate the concept, particularly in distrib amps where loads can be attached to several hundred ft of cable.