Tube Power Amp Reviews

Sidebar 3: Measurements

I measured the Sophia Electric 91-01 300B using Stereophile's loan sample of the top-of-the-line Audio Precision SYS2722 system (see the January 2008 "As We See It" and www.ap.com). I started off measuring serial no.002225 using the tubes designated for that amplifier. Something appeared to be very wrong. Fig.1 plots this sample's THD+noise percentage against output power into 8 ohms, taken from the 8 ohm output-transformer tap. While its THD+N is reasonably low at a level of a few tens of milliwatts, the Sophia clips (defined as when the THD+N reaches 1%) at just 220mW! At 426mW the distortion was 57%, and the output waveform (fig.2) was an unusual spiky signal rather than the expected flat-topped wave. In addition, the amplifier became unstable at frequencies above 20kHz.

Suspecting a bad tube, I swapped this set of tubes with those for amplifier no.002224. The behavior was the same, so I contacted Richard Wugang to let him know that I suspected the amplifiers or tubes had broken during their journey from Art's place to mine. (The front panel of 002224 had suffered some cosmetic damage, suggesting that perhaps the shipping carton had been dropped at some point in its travels. Though the pair of amplifiers is packed with a professionally made interior "clamshell" of Styrofoam, there is insufficient packing material, in my opinion, to prevent them from banging into one another under extreme conditions.)

To my surprise, once he'd received the amplifiers and set them up on his own test bench, Wugang found nothing wrong. He sent me photographs of his setup, along with measurements indicating that the samples I had tested would put out around 8W into 8 ohms from the 8 ohm tap, with a distortion reading of close to 3.5%, mostly second-order in nature.

One thing I noticed from Wugang's photographs was that the dummy load he was using was referenced to ground via the oscilloscope probe. He confirmed this and told me that the secondary of the 91-01's output transformer was not connected to the amplifier's signal or chassis grounds, which might explain my testing results. When I test an amplifier, I connect its outputs to my high-power dummy load and then connect a shielded monitor link to the Audio Precision analyzer's input. This input floats with respect to ground, as this is closest to how an amplifier will be used, loaded by a speaker that has no reference to ground. But with the Sophia, the analyzer's input needs to be separately referenced to ground for measurements—something very unusual, in my experience.

I asked Wugang to return the amplifiers to me so that I could continue testing them with a short patch cord connecting each amplifier's negative output terminal to the Audio Precision's input ground terminal. This he agreed to do.

However, when the amplifiers returned, they turned out to be different samples from those Art had reviewed, with serial nos. 002248 and 002249. Not only that, but they were physically different, with a single 8 ohm output-transformer tap rather than 4 and 8 ohm taps, a different main power-supply capacitor, no ground-lift switch on the top panel, and the hum-reducing trim pots in different places.

Time was running short. I agreed to perform my measurements on these samples, but to forward them to Art when I was done, so he could listen to them and report on any differences in a Follow-Up.

I measured no.002249, letting it warm up for an hour connected to my dummy load, but with no signal and with the load connected to the analyzer's input ground. (From here on, unless specified otherwise, all my comments refer to no.002249.) At the end of that time I adjusted the trim pots, as instructed in the manual, to reduce the hum level, working with each in turn, then repeating until any adjustment in either direction increased the noise level, which was around 700µV. Fig.3 shows the spectrum of the amplifier's output with a 1kHz tone driven at 1V into 8 ohms. Though a regular series of power-supply–related spuriae can be seen, these are mainly at a low level. The 120Hz component is the highest, at –70dB ref. 1V, with 60Hz 3dB lower. The wideband, unweighted signal/noise ratio, ref. 2.83V into 8 ohms, taken with the analyzer's output connected to the amplifier (footnote 1), was a fairly good 73.9dB, this improving to 88.9dB when A-weighted.

The voltage gain into 8 ohms of this sample of the 91-01 was 25.3dB, this a little lower than no.002225, which offered 27.7dB from its 8 ohm tap and 26.2dB from its 4 ohm tap. The new Sophia inverted signal polarity; the earlier samples were non-inverting. The input impedance of all samples was extremely high, at >150k ohms at low and middle frequencies, dropping slightly but inconsequentially at the top of the audioband. Its output impedance was also high, at 5.6 ohms at 20Hz, 3.7 ohms at 1kHz, and 8–9 ohms at 20kHz. (The earlier sample's impedance was a little higher in the bass and midrange, a little lower at 20kHz.) This high output impedance gave rise to large, ±2.1dB variations in frequency response (fig.4, gray trace) with our standard simulated loudspeaker. This graph also reveals that no.002249 rolls off in its top octave, more so with 4 ohms (red trace) than with 16 ohms (magenta), and that there is an ultrasonic resonance at 80kHz that grows more developed as the load impedance increases. However, the extended low frequencies result in an excellent 1kHz squarewave response (fig.5)—this is indeed a high-quality output transformer—though the rolled-off highs slow the risetimes of a 10kHz squarewave (fig.6).

With its output ground referenced, the 91-01 clipped—1% THD—at 3W into 8 ohms (4.8dBW, fig.7), with 6.1W available at 3% THD (7.85dBW) and 7.5W at 5% THD (8.75dBW). The amplifier gave less power into 4 ohms (fig.8) and 16 ohms (fig.9). This behavior is pretty much what I would expect from a single-ended amplifier using a 300B tube without loop negative feedback, but it's fair to note that the THD+N at powers below a few hundred milliwatts is respectably low.

Plotted against frequency at 1V into 8 ohms, the 91-01's THD+N percentage was fairly low (fig.10, blue trace), though it increased significantly into 16 ohms (magenta), and even more into 4 ohms (red). Fortunately, the distortion is predominantly the subjectively innocuous third harmonic (fig.11), though the second harmonic is the highest in level at low frequencies (fig.12) and into lower impedances (not shown). Though the 1kHz difference products with the high-frequency intermodulation test (fig.13) lay at –50dB (0.3%), on this test the Sophia actually performed much better than I had anticipated.

I repeated some of the tests floating the 91-01's output terminals and got the same poor performance I'd found with the earlier sample (see figs.1 and 2). I repeated these tests with my vintage Audio Precision System One and got the same results, with or without the ground wire connected. I have enough trust in Art's ears that I can't believe that he'd been listening to tens of percent worth of distortion without noticing it. So I set up a different front-end, using a Pioneer DV-578 SACD/CD player's digital output to feed test-signal data to my Benchmark DAC1 used as a D/A preamplifier. First I used the dummy load with the ground connection to the Audio Precision SYS2722 and got the same results shown in figs.3–12. Then I replaced the dummy load with a Polk 7A loudspeaker I'd been testing for Ariel Bitran. That way, I could listen to what was going on as well as watch it on the Audio Precision and oscilloscope screens.

I played a 1kHz tone at 0dBFS and adjusted the Benchmark's volume control to give an output from the 91-01 of 2.83V into the Polk. Then I cautiously disconnected the ground wire. The oscilloscope that was connected to the Audio Precision's monitor output immediately showed the dreadfully distorted waveform, the Audio Precision indicated 90% THD+N—and I heard no difference! Fig.14 shows the waveform of the 1kHz tone reproduced by the Polk when driven by the Sophia as picked up by a microphone; although the test gear was going crazy, you can see that it's a reasonably good sinewave, with the third harmonic lying around –48dB according to FFT analysis. My ears weren't lying to me—the test gear was! (I suspect that without the separate ground connection, the floating output of the Sophia amplifier overloads the Audio Precision's input circuit, and that overload is the source of the measured distortion.)

It appears, therefore, that Richard Wugang is correct: Although a ground reference is not required for the output-transformer secondary when the Sophia Electric 91-01 300B is used to drive a loudspeaker, that connection is necessary when measuring the amplifier in order to correctly characterize its performance. I have never encountered anything like this before. I can only suspect that there is something idiosyncratic about the 91-01's internal ground topology, especially as its noisefloor rose when the input was shorted.—John Atkinson

---

Footnote 1: Usually I use a shorting plug when measuring an amplifier's noise levels, but this increased the Sophia's noise level by about 6dB compared to when it was referenced to the Audio Precision's output ground.—John Atkinson