MJR7 Supply Rejection and Crosstalk

Supply Breakthrough

For these tests the amplifier board is mounted on a metal base, but with the top open, as shown in a photo on the MJR7 page. In a complete enclosed case external interference pickup should be lower, but transformer effects may not be reduced much, if at all. For all tests each channel had a 7R5 load resistor connected.

First of all, to show the scale of the problem, here is the spectrum of the 60V supply line when driving just the quiescent current of the two channel amplifier, i.e. about 250mA. The dB levels are relative to 316mV ( The -10dB level was calibrated using a 100Hz 100mV source ), so the 100Hz component at -14dB is 63mV. The full-wave rectified supply has its greatest ripple component at 100Hz, as expected, i.e. at twice the 50Hz UK supply frequency. The series of higher frequency components are at the other even harmonics, 200Hz, 300Hz, 400Hz etc.

The next spectrum is for the amplifier output with a shorted input, and using long supply leads so that the separation between amplifier and supply transformer is about 60cm. The levels are again in dB relative to 316mV, so the 50Hz component at -103dB is -136dB relative to the 15V maximum output level at 30Watts. Very little can be seen at the dominant 100Hz supply ripple frequency. A very small peak at around -113dB is too near the noise level to be accurate, but compared to the supply voltage 100Hz level of -14dB gives a supply rejection of about 99dB at this frequency. Moving the amplifier around changed the 50Hz level, showing that it was being picked up from the external electromagnetic field from various sources in the room, not direct from the 60V supply rail.

Next the test was repeated with the amplifier board placed about 5cm directly above the toroidal mains transformer. The 50Hz component rises to -119dB relative to full output. This is still not too bad, but in addition a series of odd harmonics at 150Hz, 250Hz etc. appear above the noise level. With a good separation between amplifier and transformer there should be no problem.

To check the effect of source impedance I repeated this test with an open-circuit amplifier input:

Clearly there is much greater effect, so if the separation between amplifier and transformer is fairly small, or there are other similar sources of interference, then it helps to keep the input signal source impedance low.
The peak at 15.6 kHz is the line frequency of a nearby TV.
It is found that the background thermal noise level actually falls compared to the previous results with a shorted input. For an inverting amplifier an increased source impedance reduces closed-loop gain which can reduce output noise.

While thinking about interference pickup I decided to repeat the previous test to determine the effect of using an input capacitor of much greater physical size. The one I tried is polypropylene, and measures 3.3 x 2.1 x 1.1 cm. I have previously suggested that it is a bad idea to use such large components, and always recommend the smallest possible physical size for the input capacitor. The next spectrum shows why. All the interference component levels are higher, typically by 12dB. If anyone really wants to use such a big capacitor it will need a low source impedance and better screening for good interference rejection.

Modifying the Supply

The supply used was just a conventional toroidal transformer with a bridge rectifier made from four 1N5402 rectifiers, feeding a parallel pair of 4700uF 63V Panasonic TSAH electrolytics. There are various methods I have seen suggested for reducing supply effects, including the use of soft recovery rectifiers and adding small capacitors in parallel with the rectifiers. To check how far such methods improve supply rejection I first repeated the test with the amplifier close to the transformer and with a shorted input, and then added 100n polyester capacitors, one in parallel with each of the rectifier diodes. These two results are shown next:

There is clearly a significant improvement with the capacitors added. Next I substituted four SBYV28-200 'soft recovery' rectifiers, and again tested with and without parallel capacitors:

The soft recovery diodes make little difference, and with added capacitors there is less improvement than before with the standard rectifiers. Just adding parallel capacitors to the 1N5402s is the best of the options tried. There are of course many alternative variations, so this result is unlikely to be the best possible.


The following tests are with one channel driving 10V into a 7R5 load, and measuring the undriven channel output. The undriven channel has a shorted input. The following results are levels in dB relative to 316mV, so relative to the 10V output level there is an additional 30dB to be added to the rejection ratio.

Only the 1kHz and 20kHz results are shown. The rejection is about the same for lower frequencies as for 1kHz, and there is a reduction at higher frequencies.

Looking at the 1kHz result, again with dB levels measured relative to 316mV, the 1kHz crosstalk is at -100dB relative to 10V. There are also components at 2kHz, 4kHz and 6kHz. With a 1kHz output the supply will have a half-wave rectified ripple component, which includes a series of even order harmonics, so that is one possible source at those frequencies. The component at 2kHz is -127dB relative to 10V. At 20kHz the crosstalk is -80dB relative to 10V, and the 40kHz crosstalk distortion component is at -110dB.