Attenuators between a tube amp and the guitar speaker: some measurements and theory

[Mike Sulzer]

Another marketing triumph. It's like the Toyta MR2 was a disaster in France as "merde"sounds like MR2 and means "sh^t".

Sorry Mike - I promise not to divert you excellent thread again.

I appreciate your comments; so please continue whenever you want to.

[Helmholtz]

Just a suggestion. Before putting humongous effort into reproducing every nuance that you can measure might it not be a good idea to figure out which ones actually have an audible effect? I bring this up as I was particularly struck by olddawg, a regularly gigging musician on the road and in the studio, who finds a simple resistive attenuator quite satisfactory.

How could we know which details have an audible effect without being able to reproduce them?
I am an experienced player myself and never got satisfying results with my resistive Tom Scholz Power Attenuator. The modified PB 100 with speaker impedance emulation is a huge improvement.
There are sound samples/comparisons of resistive and reactive dummy loads in the TGP thread linked somewhere above.

[Chuck H]

I think his point is that, while it's all well and good to search for the nuances of the operational parameters, how much does each one matter beyond bragging rights if there's no operational safety or audible advantage. So maybe start with a simple resistor and then add the reactive elements one at a time, testing by ear in between additions. Then chase the finite alterations to those elements by making those changes one at a time and ear testing between alterations.

Of course the other way would be to just test a resistive load vs. a reactive load like the Aiken circuit or the modified version of that which I use. Then, if the reactive load sounds better (which it does BTW), test the reactive load against the idealized reactive load that has been chased in this thread. If the idealized reactive load sounds better, and it may, then that's all we would know. We could not know if what idealized parameters were responsible. Only that collectively it is better. That would mean that building, or indeed manufacturing such a thing included unnecessary effort and expense. If that were part of the goal.

[Helmholtz]

Yes, you would need enough steel so that it would not saturate. On the other hand, the rod would have an effective permeability of under 10, and so I think you would need a lot more turns. I am not sure which weighs less: the steel shielding box or the steel toroid.

I have some 1 mH (I think) coils intended for crossovers, air core. Maybe it would be easiest to make a "Chuck H" sandwich, and then experiment with adding more metal until the eddy current loss is about right.

You could saw a slot into the washer for an air gap and use more turns.

The steel box was meant as housing for the complete dummy load. That's what I use.
The remark about weight I do not understand. The heaviest parts are the big choke and capacitor(s) in the low resonance tank circuit anyway.

As you don't need much more than 1mH at middle frequencies in total, 1mH air core chokes + steel core will give to high inductance. It might be possible, though, to tune the resulting inductance to a low enough value by varying the distance of 2 inversely wound, series wired chokes on a common steel core.

But maybe you not did not mean steel, when you said "adding more metal"? I made some experiments with aluminum and copper at air core coils, but the results were not satisfying as the achievable increase of apparent series resistance Rs with frequency was too small. I even tried a shorted heavy gauge "secondary" winding. The strongest effect had a massice aluminum core.
Steel cores worked much better to produce high frequency equivalent series resistance. The effective permeability of my steel core was around 2.

[nickb]

How could we know which details have an audible effect without being able to reproduce them?
I am an experienced player myself and never got satisfying results with my resistive Tom Scholz Power Attenuator. The modified PB 100 with speaker impedance emulation is a huge improvement.
There are sound samples/comparisons of resistive and reactive dummy loads in the TGP thread linked somewhere above.

Compare the speaker with the attenuator using an ABX test would seem to be the way.

I was particularly thinking of the smaller effects like eddy current and skin effect, especially as those happen at higher frequencies which the average guitar speaker cannot reproduce. I'm not saying it isn't an interesting academic exercise, but from an engineering perceptive why spend massive effort and cash on something you can't detect?

[Chuck H]

Just an observation... The physics of designing such finitely tuned, perhaps even humbucking and response accurate inductors for the project is beyond my skills by a pretty good stretch, but...

Isn't the matter compounded greatly if multiple impedance options need to be considered? With new and specific designs for each impedance. A good argument for Nick's position of trying to ID those elements responsible for relative goodness before applying unnecessary effort.

[Helmholtz]

If the idealized reactive load sounds better, and it may, then that's all we would know. We could not know if what idealized parameters were responsible. Only that collectively it is better.

The complete optimized dummy load will consist of a series wiring of the bass resonance circuit, the (high power) DCR resistor and a frequency dependent HF choke assembly (in principle similar to what you showed, but wired directly to the amp's output). It's easy to switch off (short-circuit) single components or switch between a simple HF choke and the special assembly, different resistors etc. to study their individual influence on sound. So I don't get your point.

[Helmholtz]

Multiple amp impedances can be adressed by an impedance matching (auto)transformer between amp and dummy load.

It's all not complicated science nor is it really much effort. For me it's actually a lot of fun to apply a little physics to improve my sound. As said, identifying relevant elements requires to have them available. If a simple resistive load would do, I would use it and be fine.

[Mike Sulzer]

I made some experiments with aluminum and copper at air core coils, but the results were not satisfying as the achievable increase of apparent series resistance Rs with frequency was too small. .

Exactly what I learned last night.

[Mike Sulzer]

Compare the speaker with the attenuator using an ABX test would seem to be the way.

I was particularly thinking of the smaller effects like eddy current and skin effect, especially as those happen at higher frequencies which the average guitar speaker cannot reproduce. I'm not saying it isn't an interesting academic exercise, but from an engineering perceptive why spend massive effort and cash on something you can't detect?

The plots in the first post show some increase in the real part at 3KHz. It is larger at 5KHz, of course, and some guitars and speakers have response at that frequency.

[SoulFetish]

Just a suggestion. Before putting humongous effort into reproducing every nuance that you can measure might it not be a good idea to figure out which ones actually have an audible effect? I bring this up as I was particularly struck by olddawg, a regularly gigging musician on the road and in the studio, who finds a simple resistive attenuator quite satisfactory.

This is definitely a good point to consider. But, which resistive attenuator network? For instance, as an experiment, I built a 8Ω/-6dB bridged T attenuator so I could drive my amp a bit harder in my shop. I had never used a bridged T before, but when I used for the first time I thought I had found the answer to the perfect attenuator. I mean, it sounded great. After talking it up to a friend who brought his single ended homebrew amp over, I told him to try it out. So we plugged it inline and it sounded.... blah. It really kind of made the output lifeless, and I was not expecting that at all.
To your point, Nick, I'm wondering what the main contributing factor(s) is? Could it be the push-pull pentode output stage vs a parallel beam tetrode single ended output? The reflected load impedance of 10k2 on 2XEL84s? Damping effect on the full range Jensen P15L?

[Helmholtz]

I would think that different speaker damping is at least part of the explanation. Bridged T attenuators provide impedance matching between source (amp) and attenuator+speaker as well as between amp+attenuator and speaker. But in most cases amp output impedance and speaker impedance are not equal (matched) as shown by the numbers in my post #5 above. An amp with a higher output impedance shows more "print-through" of the frequency-dependent speaker impedance on the frequency response. This means that the same speaker will produce a different frequency response when combined with amps of different output impedances.

This type of attenuator changes the relation between speaker impedance and source impedance (= damping factor) especially with amps having a high output impedance.The result is a flatter, less lively sound. You may try to wire additional resistors in series with the speaker, but this will increase attenuation as well.

[Mike Sulzer]

The capacitor for this device is important because it is used in a resonant circuit and so needs to have high Q. Films of the required size are expensive. One approach is to use a so called non-polar electrolytic. If you make your own from two polars, you can use approach 3 from the following document. I think approaches 1 and 2 are not good enough. (I did not find this document until just now, but had already arrived at the conclusion that 3 is needed. This is a better explanation than I would write.) The formatting did not work out well; method three feeds a voltage to the junction between the two through a resistor.

(https://www.flippers.com/captest244.html#ctnpo)

Reproduction of this document in whole or in part is permitted if both of the following conditions are satisfied:

1. This notice is included in its entirety at the beginning.
2. There is no charge except to cover the costs of copying.

..............................
.............................
Making Non-Polarized Capacitors from Normal Electrolytics

You may find non-polarized electrolytic capacitors in some equipment - usually TVs or monitors though some turn up in VCRs and other devices as well. Large ones may be found in motor starting applications as well. These usually do need to be replaced with non-polarized capacitors. Since polarized types are generally much cheaper, the manufacturer would have used them if it were possible.For small capacitors - say, 1 uF or less - a non-electrolytic type will very likely be satisfactory if its size - these are usually much larger - is not a problem.
There are several approaches to using normal polarized electrolytic capacitors to construct a non-polarized type.
None of these is really great and obtaining a proper replacement would be best. In the discussion below, it is assumed that a 1000 uF, 25 V non-polarized capacitor is needed.
Here are three simple approaches:

1. Connect two electrolytic capacitors of twice the uF rating and at least equal voltage rating back-back in series: - + + -
   o----------)|-----------|(-----------o
   2,000 uF 2,000 uF
   25 V 25 V
   
   It doesn't matter which sign (+ or -) is together as long as they match.The increased leakage in the reverse direction will tend to charge up the center node so that the caps will be biased with the proper polarity. However, some reverse voltage will still be unavoidable at times. For signal circuits, this is probably acceptable but use with caution in power supply and high power applications.
2. Connect two electrolytic capacitors of twice the uF rating and at least equal voltage rating back-back in series. To minimize any significant reverse voltage on the capacitors, add a pair of diodes: +---|>|----+----|<|----+
   | - + | + - |
   o-----+----)|----+-----|(----+------o
   2,000 uF 2,000 uF
   25 V 25 V
   
   Note that initially, the source will see a capacitance equal to the full capacitance (not half). But very quickly, the two caps will charge to the positive and negative peak values of the input across the combination via the diodes. In the steady state, the diodes will not conduct at all and therefore it will be as though they were not in the circuit.However, there will be some non-linearity into the circuit under transient conditions (and due to leakage which will tend to discharge the capacitors) so use with care. The diodes must be capable of passing the peak current without damage.
3. Connect two capacitors of twice the uF rating in series and bias the center point from a positive or negative DC source greater than the maximum signal expected for the circuit: +12 V
   o
   |
   /
   \ 1K
   /
   - + | + -
   o----------)|-----+-----|(-----------o
   2,000 uF 2,000 uF
   35 V 35 V
   
   The resistor value should be high compared to the impedance of the driving circuit but low compared to the leakage of the capacitors. Of course, the voltage ratings of the capacitors need to be greater than the bias plus the peak value of the signal in the opposite direction.

[Helmholtz]

A frequency dependent inductor with steel core:




It's based on a 0.33mH air core coil having a DCR of 0.3 Ohm.
Its values are Ls=0.63mH/Rs=2.1 Ohm @1kHz and Ls=0.37mH/Rs=9.4 Ohm @10khz.
Not quite there but promising. Will try a 0.39mH air core coil next.
My target values (based on the impedance of my old Celestions) are: Ls= 0.75mH/Rs= 2 Ohm @1kHz and Ls=0.4mH/Rs=15 Ohm @10kHz.

[Mike Sulzer]

A frequency dependent inductor with steel core:

[ATTACH=CONFIG]51182[/ATTACH]


It's based on a 0.33mH air core coil having a DCR of 0.3 Ohm.
Its values are Ls=0.63mH/Rs=2.1 Ohm @1kHz and Ls=0.37mH/Rs=9.4 Ohm @10khz.
Not quite there but promising. Will try a 0.39mH air core coil next.
My target values (based on the impedance of my old Celestions) are: Ls= 0.75mH/Rs= 2 Ohm @1kHz and Ls=0.4mH/Rs=15 Ohm @10kHz.

I have decided to use the two inductor approach that you described due to Zollner. I have .45mH air cores only. If you adjust the values of the Rs across them, you can get good response up to 5 KHz for the full impedance and good to 20 KHz for the magnitude. This is with a simulation; I have not measured the circuit yet.