panson_hk

The THD performance of this board can be improved by separate the input/NFB ground from the decoupling grounds. This is a common mistake in many kits.

Simply cut the trace as shown in the figure and connect the input/NFB ground to supply ground via a separate wire. THD improves as shown in the figure.

The board is ready for test. The bias level is adjusted via monitoring the THD level as well as the distortion residual.

THD vs power for 20 Hz, 1 k and 20 k for 8 Ohm are shown in the figure. It is so high compared to all my 498xx amps. For 1 k, we should have THD lower than 0.0008 %.

Let test the THD vs power without load. Without load connected, output stage contributes very little distortion. This can be used to check whether the 49810 functions as spec or not.

The amp is assembled and ready for testing. It is designed to have two rails: one for LME49810 (suggested +/- 60 V) and one for driver/output (suggested for +/- 45V). The speaker protector uPC1237 also has an input for detecting the presence of AC (half-wave rectified).

In my initial test, one rails supply to 49810 and the output stage. The positive is also connected to the AC input of upC1237.

This kit is available in many sellers in eBay and TaoBao. My one is purchased from TaoBao. The output stage does not have protection. All the resistors are pre-inserted. What a nice service to save some time.

The 49810 heatsink screw holes are not aligned well. I also worry about its size.

The Miller compensation cap is a pre-soldered chip cap on the bottom side as shown in the picture. Fuses are pre-installed in the fuse holders.

Output are five pairs of 5200/1943.

I briefly measured the LM3886 x 3 parallel board this morning. THD vs freq for one, two in parallel and three in parallel are measured.

Load of 4 Ohm was also tested. When using one 3886, the on-chip protection activates periodically when doing low frequency (~ < 500 Hz) test. No such protection operation happens for the same output level for higher frequencies (>1kHz). The problem goes away when using 3886s in parallel.

More systematic tests will follow.