Tube Amp Output into Resistive vs Speaker Loads.

[nevetslab]

Having recently dug out my Mesa Stereo Simul-Class 2-90 Power Amp to have a fresh look at it, I found while looking at the output of the amp driving a matching resistive load, that the voicing filters (Deep and Modern) didn’t seem to respond much on the audio analyzer. Yet, when driving my Ampeg BXE-115HL4 test speaker with pink noise, it clearly made a lot more difference. Manually sweeping my oscillator in the low frequency range, I was seeing a lot more output. But of course I would! It’s a transformer-coupled amp, ya dummy!

I haven’t yet set up to run an impedance plot of the test speaker, but, just to see the difference, I set up my simple Vellman PCSGU250 Bode Plotter, using my Amber 3501a Audio Analyzer to receive the amp output and a HP 353A transformer-coupled Step Attenuator for control of the generator’s output, I proceeded to generate Frequency response plots showing the baseline 4 ohm Resistive load plot , first showing the two Voicing Filters (Deep & Modern), and the Min, 50% and Max settings of the Presence Control on this amp.

Then, I ran the same baseline plot of 4 ohm Resistive @ 50% Presence and added the response plots with the Test Speaker attached. What a world of difference that shows!

For example, when I added the Deep Voicing Filter, the LF output where the speaker hits maximum impedance at around 65Hz, the output is 15dB above the 4 ohm resistive load level. The Presence range, which is now boosting in the HF region where the inductive reactance of the speaker is again high impedance, it’s a good 12dB up from the baseline.





That old comment stating that tube amps just seem to sound louder than solid state amps….I’d say so. The results are certainly more dramatic under speaker drive conditions than it is driving a dummy load.

Pardon my typo's in the annotation of the graphs. No simple way to recover from that.

[nevetslab]

Still to come in this series are the acoustical measurements, using a solid state power amp, set for the same output level @ 160Hz 1/3 Oct Pink Noise and the same level from the Mesa Stereo Simul-Class 290. 160Hz is about where the speaker is at minimum impedance and zero phase. Once I've set those levels up, then I'll use wide band Pink noise and record it on a Bruel & Kjaer 2035 Spectrum Analyzer (1/6 Oct). I'll also run an impedance plot of the Ampeg BXE-115HL4 bass speaker that I use as one of my shop test speakers.

[Leo_Gnardo]

Bruel & Kjaer 2035 Spectrum Analyzer (1/6 Oct).

I had a feeling you were going to break out that B&K. Nice to have the gear & time to analyze to the nth degree!

[nevetslab]

I meant to post the schematics for the Mesa Simul-Class 2-90. Shown below:



I'm also curios to see current waveforms of the two different power tube pairs in this circuit. I'll break out the Tek current probe for that, having been following the lengthy thread on Randall's Supro Project "why are these cathode based output tubes drawing so much?

And yes....nice to have access to that gear. it's been a enjoyable day in the physic's lab.

[nickb]

Do you have a calibrated mic? It would be interesting to see the speaker voltage plotted against SPL.

[teemuk]

Amps with high Z output (which isn’t necessarily just a tube amp characteristic) will create higher voltage potential across terminals of higher impedance load than across terminals of lower impedance load.

In essence, such amplifier has higher voltage gain at higher impedances than at lower impedances, such as speaker’s “nominal impedance”.

When you test a purely resistive load against a reactive load this characteristic will manifest itself somewhat clearly. A reactive loudspeaker load, for example, has some typical characteristics: Impedance increases towards higher frequencies, where inductance of the voice coil starts to become significant parameter. Another impedance peak appears at speaker system’s “resonant frequency”, which is usually around the low end limit of the speaker*. At other frequencies the load impedance is closer to “nominal” value. So, when such load is hooked to an amp with high-ish output Z the amplifier introduces more voltage gain at frequencies at which those impedance peaks are located. That is, towards highest and lowest frequencies of the effective bandwidth.

*The Ampeg's response actually shows two peaks. This indicates that the load is actually not just a loudspeaker but a tuned system, IOW a "reflex cabinet". The reflex structure introduces another resonant peak - in opposite direction to driver's own "resonant" peak - and which - when properly tuned - is at the same frequency as well. One resonance attenuates the other one, and the remaining of that attenuation are those two peaks you see in the lower bass response. They are much lower in magnitude than the single resonant impedance peak you'd have at driver's resonant frequency would be. Yes, also the cabinet matters in many occasions!

Ok, so we add negative feedback-based tone controls to such setup. Negative feedback, in general, will decrease effective output Z, which in turn makes frequency response more “level” to loads with widely varying impedance. The aforementioned tone controls, on the other hand, work by (frequency selectively) varying magnitude of negative feedback. In extremes the control adjusts gain between open loop gain of the amp (no negative feedback applied) and closed loop gain (negative feedback applied). Range of the control is approximately difference of the figures: OLG - CLG.

When you plot the response with a resistive load the load impedance is usually selected to be in the order of speaker’s nominal impedance. For example, for “8-ohm speaker” (nominally) a corresponding purely resistive load is 8 ohms.

…BUT – and here comes the important part - the load impedance IS NOT equal to nominal impedance at the frequencies at which those tone controls operate. The load impedance at those frequencies is HIGHER than the nominal load impedance. The amplifier will therefore correspondingly also produce higher voltage output to such load than it will produce to a load equal to nominal impedance.

This is why the controls have a great deal more range with typical reactive loads than with a resistive load.

The same characteristic also turns the response into something different than the “shelving” response the tone controls provide with a resistive load. The response will, in fact, reflect the peaks in the load impedance.

That old comment stating that tube amps just seem to sound louder than solid state amps….I’d say so. The results are certainly more dramatic under speaker drive conditions than it is driving a dummy load.

This logic, however, fails to the simple fact that SS -guitar- amps are usually made to have high-ish effective output impedance, similarly to tube amps. This has been usual design approach for a couple of decades already. They will likewise display similar non-linearity in the response with reactive loads so what comes to this characteristic is pretty much apples to apples comparison.

[J M Fahey]

Let me start by THANKING Nevetslab for his work, taking care to experiment and register everything, and posting here

As of the findings, I agree and add, in my own words, my simpler explanation/example:
* if by use of any filter you increase some frequencies fed to the amp by 6 dB, output will show an increase of 6 dB.
Applies to SS amps and to tube amps with flat resistive load.
* if said filter is not an "external" one (in the preamp or between preamp and power amp) but an *internal* one , boosting by *reducing* NFB selectively, you will have *more¨* than 6 dB increase at frecuencies where load impedance is higher than nominal, a.k.a. "Speaker Loads"
Said increase could easily be twice as much, or close to 12 dB ... which is clearly seen in the graphs.

So what´s the real difference (in this case) between Tubes and SS?

* "Normal" Tube configurations are inherently high impedance, source being pentode plates, almost perfect current sources .
That they are transformer coupled to the load keeps that characteristic working.
And when you "let them do their thing" , as in lowering or omitting NFB, their output impedance rises a lot.

* "Normal" SS amps are inherently low impedance , coming from transistor Emitters , be they actual emitters (Darlingtons) or "created" emitters: collectors which behave like emitters in a quasi complementary transistor pair, end result is the same.
So if you add Presence and Resonance circuits in the NFB loop, they will boost just the "plain" value you added, but are not "leberating" any "wild high Impedance Demon hidden in the bottle" .... because there never was such a "demon" there to begin with.

Hence Presence and Resonance effect will be less dramatic in standard (Lin type) SS power amps.

Now, if you have a transformer driven SS power amp , where output now essentially comes from collectors, wilder things happen when overdriven.

[Leo_Gnardo]

Do you have a calibrated mic?

Yes it sure would be good to see a nice B&K "instrument" mic in use with the other super hi quality B&K test gear.

Also, having nothing to do with measuring the amp/speaker, it's always bothered me how Mesa set up their cooling system in that amp. The tubes nearest the fan benefit the most. Tubes further away, not so much. A couple of my crustomers have/had these, output tube life generally a year or two at best in spite of the fannage. One more opportunity to wonder "at Mesa, what were they thinking?"

[nevetslab]

First up....yes indeed I have calibrated Bruel & Kjaer mics...several types. 4133 Free Field, 4134 Pressure Response, which is what I'll be using here in the shop, both being 1/2" size. 4144 1" Pressure Response & 4145 1" Free Field, 2619 preamps, and all the cabling. The various Measuring Amps and Analyzers I have all have the front end for powering their mics.

What we see in the Voltage Output responses from the Mesa amp driving the Ampeg BXE-115HL4 speaker is the 30Hz Port tuning, where the response dips down to near nominal 4 ohm impedance, as it also down in the low-midrange trough, which is where the 'nominal' speaker impedance is specified. Sweeping from 10Hz, we get to see both high impedance peaks reflected. If I had selected 1/24Oct resolution it would have been much smoother than the 1/6 oct resolution used in the sweep. We're seeing a reflection of the speaker system's impedance, though it is weighted a bit by the actual output impedance of the Tube amp. Once I get the next test set-up configured, so I can plot the speaker system impedance, I'll also do a plot of the output impedance of the Mesa amp, driving the 10mA constant current into the output terminal of the amp. I haven't tried this on a tube amp, and hope this doesn't cause any harm to the amp. Using that technique to measure the output impedance of a SS power Amp, I'd normally use 100mA so the there's a greater voltage potential to read, as you're typically seeing low milliohms of power amp's source impedance, which usually rises with increasing frequency, both in part to it's gain-bandwidth product and the RF output network, usually consisting of an aircoil inductor in series with the output, a resistor across it, then a resistor/capacitor to ground.

With this Mesa simul-Class 290, the Voicing Filters and, of course, the Presence control are all part of the negative feedback loop. So, as we see in the voltage output response graphs, their effect is greatly magnified by the loudspeaker's load impedance. If they were upstream, ahead of the power amp stage, then thre wouldn't be any change in their response.

I very much llike 'teemuk's discussion on the workings of control networks in the NFB loop, as is typical in this Mesa design.

I'll have to look thru our inventory of SS amps to see what we have that IS on the NOT-SO-LOW Output Impedance design. What I have in the shop are traditional sound system type power amps (Crown DC300A, BGW GTC, Alesis RA100, Yamaha P-2100. What comes to mind as a good example of a highish-output-impedance SS amp?

I agree with Leo regarding the location of the fan. Not to mention I've more than once been bitten by the exposed fan blades!! The tubes at the opposite end of the tightly-packed 8 tubes get lots of hot air blown on them. That's a task for later mechanical engineering, as time and desire permits.

[nevetslab]

In attempt to make use of the PCSGU250 computer-controlled Bode Plotter that I used in the Freq response measurements, I was going to modify the set up to create a Network Analyzer. Normally, I'd use the B & K plotter with the generators & analyzers. But, being out of fresh plotter pens for my B & K XY Plotter, which will drive my 1023 & 1027 generators, both having Compressor loops in them so you can generate constant current from the generator output for making the impedance plots, I cobbled together a similar system using a stand-alone Compressor (part of their 1405 Noise Generator box). Nothing was working correctly, and just found I've a problem with the stand-alone compressor...essential in the measurement system to yield constant current from whatever sweep generator I use. So, I gotta take a time out to read thru the service manual and figure it out. Sigh....

[J M Fahey]

it's always bothered me how Mesa set up their cooling system in that amp. The tubes nearest the fan benefit the most. Tubes further away, not so much. A couple of my crustomers have/had these, output tube life generally a year or two at best in spite of the fannage. One more opportunity to wonder "at Mesa, what were they thinking?"

Fans do not cool *tubes*
To be more precise, all they can cool is the glass envelope, definitely not the plates which are insulated by vacuum, the best heat insulator in the Universe and the secret behind Thermos flasks, go figure.
What you cool is the amp guts, from chassis to transformers to components, which of course is not a bad thing.
Plate cooling mechanism is by radiation, not conduction.

[teemuk]

Yes, plates dissipate heat by radiation. The envelope, on other hand, by convection. Radiation is more effective the higher the temperature difference is, therefore a cooled envelope aids radiation. Plus it can prevent “hot spots”. In the Mesa amp the cooling is partially “forced”, but I think in this case even little airflow helps in keeping the envelopes cooler. ..and yes, removes hot air from heating stuff around.

Regarding earlier question about solid-state amps with high-ish output impedance, my suggestion is GUITAR amps, in general. Steer away from “acoustic” amplifiers, bass amplifiers, and especially “Hi-Fi” amplifiers, including PA power amps, amps within self-powered cabs and alike. In them “coloration” with EQ is not generally preferable. Bass amplifiers may exhibit such characteristic, sometimes, but in excess it makes the low end way too “flubby” and underdamped.

Look at guitar amps. Many of them utilize a very low-resistance “sampling” resistor connected to the oyjer terminal of the speaker load. This converts load –current- into voltage signal fed back to the power amp’s input as a secondary feedback path. This configuration artificially increases effective output impedance. While the loop “operates”, naturally.

[Leo_Gnardo]

Fans do not cool *tubes*
To be more precise, all they can cool is the glass envelope, definitely not the plates which are insulated by vacuum, the best heat insulator in the Universe and the secret behind Thermos flasks, go figure.
What you cool is the amp guts, from chassis to transformers to components, which of course is not a bad thing.
Plate cooling mechanism is by radiation, not conduction.

Good observation as usual Juan. However, note the position of tubes and fan in the construction of this particular model. Wind from the fan doesn't enter the chassis of the amp, can't cool transformers & interior components. Output tubes stick out the back in two rows, their bases bolted into the steel panel that forms the back panel of the chassis. Despite the fact that plates radiate heat, the tubes still depend on convection to expel that heat. I find output tubes in this model of Mesa tend to fail in the rank furthest from the fan while tubes closer to the fan survive though the glass may get covered in dust. Checking the tube temp - as well as I can - in normal operation reveals an obvious trend. I use a "laser" infrared thermometer but granted can't get a good angle to monitor temp from the tubes' sides as chassis extensions from top & bottom get in the way, so I kind of limited to reading from the tops of the bulbs sticking out the back of the amp.

If it was an old SVT, fair enough. Fans in those pointed straight at the big power transformer. Some of the wind bounces towards the output tubes and helps carry heat away in a more or less even pattern. Not at all the same case in Mesa's 90/90.

[nevetslab]

Impedance Plotting System up and running

Having made the repairs to the Bruel & Kjaer 1405 Noise Generator/Compressor Amp box that was needed for generating the constant current source to drive the test speaker, I tried to adjust the calibration of the system to match the dB vertical lines, but wasn’t able to tweak it as desired, so I plotted resistive calibration lines to scale the impedance plot of the Ampeg BXE-115HL4 Bass Speaker used with the Mesa Simul-Class 290 amp in the previous plots.



We see the Port Tuning of the cabinet between the two LF peaks at around 36Hz, then the 'nominal' impedance in the wide trough around 160hz (a bit below 4 ohms), where the impedance rises steadily on the inductive reactance of the 15" spkr, then it drops down in magnitude where the Piezo Tweeter comes in, with it's impedance then rising beyond our hearing range. I think the Piezo & crossover are added in parallel with the 15" speaker.

After fumbling with cables to get the Sense Resistor into place, I stopped to construct a simple 3-pair Dual Binding Post box with copper buss bars connecting the Load and Sense connectors in series, with the Source connectors across them, so the wiring and set-up calibration was simplified. There’s a few photos of the cobbled-together measurement system used.



A simplified Test System diagram is sown below.



The PCSGU250 is an inexpensive laptop-controlled USB scope/generator with a Bode Plotter mode that I used to generate the frequency response plots earlier. In this step, I added a Bruel & Kjaer 1405 Compressor Amplifier (part of the Noise Generator) to feed the power amp which drives the test speaker. I inserted a 1 ohm resistor in series with the Ampeg BXE-115HL4 Bass Speaker, and sampled the voltage with a Bruel & Kjaer 2607 Measuring Amp, set for 10mV FS, and the output of that feeds the Compressor Control Input. Then, the Compressor level and the Amplifier Level are adjusted to achieve that 10mA constant current. The AC Voltage across the speaker is measured differentially by an Amber 3501a Audio Analyzer, whose output feeds one of the two channels of the PCSGU250 (and the scope), while the generator output feeds the other channel.

Next step is to drive a constant current sine wave into the output of the Mesa Simul-Class 290 to measure the output impedance of the amp on both the 4 ohm and 8 ohm taps. And hope it doesn’t object to that, not having tried this with a tube amp.

Then it's on to making the SPL measurements to see what difference is found with the same voltage drive level (into the Ampeg test speaker) on both a solid state power amp vs this Mesa Stereo Simu-lClass 290. I'll set the drive level up at 160Hz, where the impedance is at the 'nominal value (4 Ohms)

[nevetslab]

Tube Amp Output Impedance Plotting Results

After first trying to drive the open-circuit 4-Ohm Output Jack of the Mesa Stereo Simul-Class 290 Power Amp from a 1k ohm signal source impedance, and had it break into a large asymmetrical square wave output and pull additional AC mains current, I quickly shut it down (S/B mode) to re-think the procedure. I had a feeling it would object with open circuit. I then tried a simple test, feeding a 1kHz oscillator input into the amp and driving an output voltage into a 4 ohm and then 5 ohm load, measuring the voltage difference, compute the output currents, and derive the difference between the two and calculate the source impedance. While this works, it would be a long tedious set of measurements.

The PCSGU250 USB scope/generator/Bode Plotter has a max output of 10V P-P (3.54V RMS). Looking at the loudspeaker impedance plot I made of the Ampeg BXE-115HL4, seeing the two LF peaks on both sides of the port tuning, being a bit over 32 ohms, I connected a 32 ohm load across the 4 ohm output jack. Then, adding 1k source resistor from the 10V RMS output of the Amber 3501a’s generator, I looked to see how the Mesa amp would behave feeding that signal into it’s output That worked. So, I re-configured the speaker Impedance Plotting set-up to use a 300 ohm sense resistor from the output of the Alesis Power Amp, connected a B & K Bridging Transformer across that Sense Resistor to feed the B & K 2607 Measuring Amp, which feeds the Compressor Control Input. I then adjusted both the power amp level control and the Compressor Control Voltage level to obtain a steady 3V RMS, and now I had my 10mA constant current source.

I attached 32 ohms across the 4 ohm Output jack, connected that to the Amber 3501a Audio Analyzer to drive the Impedance Magnitude input of the PCSGU250 and the raw generator to the other input, I took a look at the results, not yet having run resistive calibration lines to scale the plot. That seemed to work, at least enough to get in the ballpark. I disconnected the load from the power amp, (it being a decade resistor box), and ran plots of 32 ohm, 16 ohm, 8 ohm, 4 ohm and 2 ohm. Then, I plugged it back into the amp, set it for 32 ohm load, and switched the amp back into operate.

Now, there is an obvious error in the plots, that being the constant 32 ohm load (for the 4 ohm plots), loading down the actual output impedance. But, it would at least yield a trend to start with. After recording the first plot, with the Presence Control set at 50%, I saw a high magnitude resonance peak in the 15Hz region. After running two more plots, one at Min and one at Max Presence Control settings, I saw what looks to help explain why some Mesa amps go into LF oscillation (motor-boating).



I next looked at the 8 ohm tap, and decided to try 64 ohm load for that, with the results being very similar in shape to the 4 ohm plots. I then went back to the 4 ohm jack, and turned on the Deep Voicing Filter to see what that did. I set the Presence to Max, having seen that had the smallest resonance peak. I also found the Deep Voicing filter swamps out the high Q LF resonance that the Presence control yielded. The ‘Modern’ Voicing filter didn’t change the Output Impedance shape at all, so I didn’t include that plot.

With the Presence control set to Max, the output impedance of the amp is highest amongst the curves, and has the lowest LF peak, while the Min Presence control setting has the highest LF resonance peak and the lowest HF impedance, dropping below 2 ohms @ 6khz. In the next 4 ohm tap impedance plot using the Deep Voicing Filter, when the Presence Control was set to 50%, it yields lower High-Mid Frequency impedance, and just a very minor change in the LF portion of the curve. That Voicing network swamps out the high resonance behavior of the Presence network that we see when the Deep Vicing inst' engaged.

I haven't yet checked to see just how high I could raise the load resistor across the Mesa Poser Amp's output jack, knowing it will misbehave with NO LOAD. I will do some spot checks manually using the Change in Load Resistor method to compute source impedance to see just how much error these plots are showing.