The published measured performance of EHA5 is pretty impressive. Is it really that good? Let’s see what those published measurements didn’t tell us.
First is the ‘appalling’ SNR of 146dB. Topping calculated the SNR using the maximum output voltage divided by the zero-signal A-weighted noise. I was able to verify that the zero-signal A-weighted noise is within spec (measured at 79uV Gain=High). However, the raw measurement without A-weighting and AES-17 filter is not that great:
| Vol Pot position | 90k BW, no filter | AES17-20k+A-wt |
| Min | 1.07mV | 79uV |
| Mid (2:30 position) | 5.45mV | 140uV |
| Max | 0.91mV | 104uV |
The SNR degradation is most likely caused by large amount of ultra-sonic noise at the output. Once the unit is powered off, those noise spikes are gone. Apparently the noise comes from the unit itself, perhaps the switching power supplies (both the wall wart and the internal switching regulators for the negative rail and the bias).
View attachment 311222
The
frequency response looks pretty flat, except for the slight uptick near 20KHz. Did anyone wonder what it looks like beyond 20KHz? Here it is:
View attachment 311223
The Gain switch is set at High. A 10:1 attenuator is added at the input of Audio Precision unit to extend the measurable voltage range to cover the maximum output voltage of EHA-5. Adding the attenuation ratio of 10x (20dB) on top of the curve, the actual gain of EHA5 is about 45dB at 1kHz.
We can clearly spot the 6dB gain peaking at about 32.7kHz. Beyond the peak, the response drops quickly. The time-domain impulse response below proved the existence of a resonance peak, which corroborates with the poor square-wave response posted
here.
View attachment 311224
Most people noticed the increased THD in the low end (see
here). The spec shows output level of 700Vrms, which seems to surpass most direct-drive eStat amps. How do those two parameters play together? Below is THD+N vs output level at different frequencies. We can see that at 20Hz, EHA5 can only output up to 229V before the amp goes to protection, and the distortion is close to 3% already. The high end output level is also limited. In order to show the real-world distortion figures, the measurement used 90kHz bandwidth and no weighting.
View attachment 311225
Topping told us the EHA5 can output 700V, what was left out is that it can’t output 700V across the entire audio frequency range.
From the simplified block diagram, we can see this somewhat unusual low pass filter between the op amp buffer and the volume pot. Dr. KG describes the low pass filter as a slew-rate filter. I consider it having two purposes; one is to limit the high frequency component from entering the main amp (we’ll see the reason later), the second is to compensate the gain peaking of the output transformer resonance so that the frequency response is near flat up to 20kHz.
View attachment 311226
The frequency response from the amp section with the transformer disconnected looks like this:
View attachment 311235
Both curves are about -1.8dB down at 20kHz and -3dB at about 27.65kHz. In other words, the amp section is pre-eq’ed to make the overall FR look flat (up to 20kHz).
Going back to the question why EHA5 uses an opamp buffer in the front like the O2 headphone amp, and a volume pot value as low as 1k Ohm. The noise measurement at the beginning of this post answered part of the question. The main amp probably has a poor current noise density spec, which means it can only work with low source impedance in order to reach the desired SNR. The THD FFT also shows a not-so-great moment (see below). This is the amp section driving a 10 ohm load with the volume knob at roughly 1 o’clock position. The second harmonic is roughly at -90dB of fundamental. This is still decent, but not as superb as you would typically see in a Topping headphone amp measurement review. That seems to indicate that the main amp’s input stage is not that linear, even slightly larger source impedance would add quite a bit of harmonics. The silver lining here is that the gain selection is done in its feedback path, making the linearity in the low gain position better than in the high gain position. That would be a relief if you don’t need the high gain position.
View attachment 311236
Speaking of the transformer, they are slightly larger than the ones in my SRD-7SB. The difference is quite small, though. The impedance measurement of the EHA-5 transformer is shown below with its output open, at 100V and 200V driving a Stax SR-404. At the low end, the impedance drops due to the limited primary inductance. The situation gets worse due to core saturation under high output level (green curve). At the top end, the impedance seen by the amp dips to only 2 to 3 ohms at around 40kHz. That’s probably why the LPF is added to prevent large ultrasonic signals from entering the main amp.
View attachment 311237
Compared to the impedance curve of the SRD-7SB below, we can see both transformers follow the same trend. The EHA5 transformer has less impedance kinks in the low end. We can see its improvements compared to the SRD-7SB, but the inherent limitation of a transformer-based solution is still there. It explains why the top end and low end output levels are reduced compared to 1kHz. Driving large amplitude into a very low impedance load is too much to ask for a headphone amp module like the ones in the EHA5.
View attachment 311238
The takeaways of this measurement session is the following: If we only test the same parameters the manufactures test their units, under the same conditions the manufactures test them with, we can expect to get the same results that the manufactures want us to see. Occasional we catch a test escape or a unit that is out of maintenance, which doesn’t necessarily reflect the majority of the units in the market. To understand the true performance of the unit, we would need to get a bit more creative.
But yeah, it confused me for a bit before I drew out the waveform...
