Crescendo_complete_1_small.jpg  The Crescendo - tamed...  Crescendo_complete_1_small.jpg

<< back to page 1                                                                page 2


"Why do amplifiers oscillate, and are oscillators often unwilling to start?" From the beginning in 1982, the Crescendo is troubled by persistent oscillation problems. When the design is built with the standard parts, very often oscillations occur on the positive half of the signal caused by the MOSFET's and wirewound source resistors in the output stage.

In the "Elektor" of those days, it was advised to insert an extra 1nF capacitor between the source-pins of the 2SK135's. However, in a number of cases this is not sufficient to suppress the oscillations. A better solution is to use low-inductive 5W metal film resistors for the source resistors. These are very common these days, compared to their availability in 1982. Another possibility is to use five 1Ω/1W metal film resistors in parallel. This was used in the Mini-Crescendo that was released in 1984.

The version that was built on the previous page behaved itself very good for years. That all changed when the amplifier was used in another audio set: it soon appeared that the 10Ω Zobel resistors were getting very hot, a clear sign of high-frequency oscillation. Measurements with a scope showed that as soon as the speaker cables were connected, mainly the right channel started to oscillate heavily. On the testbench there was no sign of oscillations however, and a square wave was reproduced nicely without overshoot also, even with a capacitive load. One explanation could be that the rather long speaker cables work as an antenna and pick up interference, which in turn enters the amplifier through the output and makes it oscillate. The right channel has a longer speaker cable attached to it, so more interference is picked up.

Incoming interference through the output of the amplifier can be reduced by enlarging C16 in the Zobel-circuit, from 22nF to 100nF for example. The cross-over point of the network is lowered by this and shifts from 700kHz to 160kHz, so incoming noise is supressed better. Use a capacitor for C16 with good high-frequency properties and capable of withstanding high current, like an MKP capacitor.


The heatsinks of T8 and T10 are also part of the problem in this case. The SK09 heatsinks that are used here are standard 50mm and thus rather long. Through the TO-126 housing, they are connected to the middle pin of T8 and T10 (the collector) and with that to the input of the output stage: the gates of the MOSFETs. The output impedance at the collectors of T8 and T10 is high, because they are configured as current sources and so it is a sensitive point. The MOSFETs are located relatively close to T8 and T10, and the big TO-3 casing is connected to the source, which corresponds to the output of the amplifier. Between the relatively large surface of these heatsinks (the gates) and the casing of the MOSFETs (the sources) there is a parasitic capacity present, and for high frequencies the feedback from output to input is a fact. If sufficient noise enters through the speaker cables, local oscillation occurs at a frequency of several MHz. A slight touch of one of the heatsinks of T8 or T10 was all that was needed to stop the oscillations.

The heatsink of the output stage of an amplifier is automatically connected to ground most of the times, because it is fixed to the metal housing. With most designs, the transistors of the driver stage are mounted on the same piece of aluminium as the output stage, so the mentioned problem does not occur. To make sure the heatsinks no longer act as open-field 'antennas' attached to the gates of the MOSFETs, the have to be isolated from them. This can be realized very simply by mounting T8 and T10 with a mica washer or a silicon isolation pad. With the aid of an extra soldering eyelet, the heatsinks can be connected to the local power-supply ground. To realize this, drill a small hole in the PCB near C6 and C7, at the side that is connected to the ground trace, and connect the eyelets to ground with two short pieces of wire.

Cascode load

The load of the cascode drivers consists only of the input capacitance of the MOSFET's in the output stage. The impedance  of this capacity is (like with any capacitor) depending on the frequency, and thus, so is the driver current that will be running. To make this load less frequency-dependent, a 22k resistor is placed between the collectors of T8 and T10 and ground. Another advantage of this is that the gates of the MOSFET's are now connected to ground through this resistor.  The impedance at this point is lowered and with that it is less sensitive to interference.

Overview of additional measures

- use low-induction resistors for R27 to R30
- enlarge C16 of the Zobel-network to 100nF and use an MKP
- isolate T8 and T10 from their heatsinks and connect these to ground
- mount 22k load resistors for T8 and T10
- T8 and T10 may be replaced with MJE340 and MJE350
- place a 100nF capacitor across P1
- lower C2 to 100pF polystyrene

Eventually, the new schematic is as follows:

To page 3: "The DC Servo" >>