The last class D amp we looked at was a stellar reference design from TI based on the TPA3251. The reference designs from TI are especially interesting because they represent the best-case application of the part. Their engineers know what the part needs to excel.
This time, we purchased a TPA3255-based no-name class D amp that is commonly seen on E-bay or Amazon for around $60 or $70.
These offerings are usually very light on specs, but the sellers place great importance on the value and the brand of the DC bus capacitors. Curious.
The amp input was single ended (L + R channel), while the output is push-pull. If you reference Figure 29 in the TPA3255 spec, and imagine the Input B and Input D inputs grounded, this likely (but not certain) reflects the configuration used on this amp (BTL). Note that in this mode, the speakers are DC coupled to the TPA3255. There are other modes (Single Ended) where you can have 4 independent speakers using AC coupling (figure 30). Or, you can parallel the stereo outs and drive a single speaker (figure 31) in what TI calls PBTL
It's important to measure a class D amp output differentially. But this poses some challenge to ensure the gain lineup is correct. A good way to begin an evaluation is to first determine the approximate gain of the amp, and then make a few DVM measurements to ensure things are as they appear.
The QA401 was setup to generate a fixed tone at -20 dBV and that signal was injected into the amp input. The output was a four ohm load, comprised of two 2 ohm resistors. The QA401 was set to measure differentially across one of the two ohm resistors to extend the max power handling. With a -20 dBV input, the voltage measured across the 4 ohm resistor was 2.359Vrms, which is (2.359^2)/4 = 1.39W . Here is where it can get conceptually tricky: When we measure differentially, we're doubling the effective voltage swing. However, we're also measuring across half of the load. Power is related to the square of the voltage, and linearly to the load impedance. So, this is a factor of 2X we need to account for. When reading in dBV or dBu, this is handled in the context menu.
With those adjustments made and checked with the DVM, we now had the system gains set correctly and could make some measurements.
The noise floor of the amp was measured by shorting the input to the amp and measuring the output. Because the amp has a fair bit of gain (almost 28 dB), it's important to make sure that the amp input sees a low impedance termination when making this measurement.
The first measurement was made at 50V, without any A-weighting.
The red trace was the right channel input on the QA401 shorted. This measurement indicates 35-40 dB of margin to the QA401's noise floor. At 50V, there's an instability/oscillation in the output of the amp. This was verified by spraying parts of the amplifier with cold spray. When the LM2575 section of the board was chilled, the instability shifted and went away. This lets us verify the instability isn't coming from the 50V power supply and is on the amplifier board itself.
The source of the clock isn't known. The TPA3255 clocks are around 3 MHz. What is also interesting is that the instabilities area to be harmonically related and symmetric around the 50 KHz clock. That is, there's a oscillation at 50 +/- 14 KHz and 50 +/- 2* 14 KHz. What is clear is the presence of the instabilities means a one or more design flaw exists in the amplifier supplies. This should not be happening in a well designed product.
Note that the instabilities disappear as the input voltage is dropped to 30V (and the 20 to 20 KHz noise drops by to 1/3rd of the 50V level):
Since noise measurements generally specify an A-weighting filter, that plot is shown below:
TI lists the spec for single-ended noise at 51 volts supply rail at 160uV using an AES17 (20 KHz brickwall analog filter). In the evaluation of the we did previously on the TI TPA3251, we reported a noise floor of 88 uV, which was a bit higher than TI's typical spec of 62 uV on that part. This 311uV figure is roughly twice what TI typically expects for this part. Now, practically, this noise is pushing about 24 nano-watts into a speaker, which is about 76 dB below the 1W sensitivity of a speaker. So, if your speakers are 90 dBSPL per watt efficient, then this noise would be 76 dB below that, or about 13 dBSPL. That's close to silent. Most would be hard pressed to hear this hiss.
The response of the amp is plotted below. At 18 KHz, the output is down about 1.4 dB.
TI lists the BTL THD+N @ 1W as 0.006% typical in section 7.6 of the spec, and Figure 1 shows the THD+N @ 1W as 0.006%. In the plot below for 1W, the THD is indicated at 0.03%, which is considerably higher. However, the noise and distortion (indicated at -64 dB) improves by about 12 dB if the measurement ends at 10K instead of 20K. This indicates there's significant degradation from the previously noted instability/oscillation. There's also the question of the external filters TI's measurements used. TI indicates an AUX-0025 filter was used. This filter has a passband out to 20 KHz, and then drops transitions to >52 dB of attenuation above 250 KHz. So, this is roughly a 3rd order filter. At 40 KHz it might have 18 dB, and at 80 KHz it might have 36 dB of attenuation. TI also notes that they are running THD+N out to 80 KHz. But much of that additional bandwidth has little contribution due to the attenuation from the AUX-0025 filter.
The plots below are 1W, 10W, 100W, 200W and 300W of power output.
Some comments on the above: The THD improves between 1W and 10W, which is also what the Figure 1 plot from TI shows. At 200W, the THD and THD+N are quite respectable. TI indicates 315W into 4 ohms in BTL will result in 10% distortion. In the plot above, 12% was measured at 305W, which agrees with TI's result.
An IMD measurement was made. This metric isn't reported or measured by TI or the amp maker, and is offered for reference only.
In the coming months, we'll look at the impact of a lower-cost external filter for the QA401, as well as a comparison to TI's reference design board on the TPA3255 chipset. Stay tuned.