Offset_PCB_close_small.jpg  The DC Servo  Offset_PCB_close_small.jpg

<< back to page 1                                                                page 3

In april 2001, the Crescendo Millennium Edition was released. This design is a further development of the original Crescendo, in which a number of changes were introduced, notably to improve the stability. A number of these can be found on the previous page. The output power is lower than that of the old Crescendo: the supply voltage is lowered to +/-50V and two MOSFET's are used instead of four. The current through the differential input stage and the cascode was raised significantly, extra compensation was added and a DC servo was fitted. This last item is a useful piece of electronics, that can be used with the old Crescendo as well.


With the Crescendo, the offset at the output mostly depends on the differences between the components in the differential stage at the input. Badly matched transistors can lead to tens of mV offset here, which in turn leads to tens of mA DC current through the loudspeakers. Because this offset is not supposed to be amplifier as well (like the audio signal), a capacitor is fitted in the feedback loop, in series with resistor R4 (1k). In this case, a bipolar electrolytic constructed with C4 and C5. For audio signals, R4 is normally connected to ground through C4 and C5, and the amplification factor is set by the ratio between R5//R6 and R4. For DC voltages, R4 doesn't seem to exist, because C4 and C5 act like an open circuit for direct current. The offset voltage that's present at the output is not divided and fully appears at the second input of the differential stage: the feedback ratio is 100% and the amplification for DC is 1. This way, the offset is limited to several tens of mV and is not amplified 32 times.

This method works well, but has an important disadvantage: capacitors C4 and C5 are directly present in the signal path. Electrolytic capacitors produce clearly measurable distortion when an audio signal is passed through them, certainly if no DC voltage is present across the capacitor. In this sensitive part of the circuit, the feedbak loop, it's better not to use electrolytics. Of course, a Black Gate N-type can be used, like is done on the first page, but the best electrolytic is no electrolytic at all in this case. A monstrously big 100uF MKP or MKT capacitor could be an alternative... An extra advantage is that the low frequency cross-over point that is formed by R4 and C4/C5 is eliminated.

The DC servo

A DC servo sees to it that the output of the amplifier is actively held at zero volts. One single opamp can take over the job from the electrolytics. The opamp measures the offset voltage at the output of the amplifier and adjusts it to zero by applying an inverted correction voltage at the input. The offset appears to be moved to the input this way. A number of capacitors and high value resistors, placed at the input and the output of the circuit, make sure that the amplifier does not "see" the servo and the audio signal is not affected.

Inverting_integrator.gifThe non-inverting input of the opamp is connected to ground, so the incoming voltage is compared to 0V. This is the ideal target value for the final output offset. The amp's actual offset voltage enters at the inverting input, through R33. Together with C22, this forms a low-pass filter with a very low cross-over frequency. This way, audio signals are blocked and only the DC voltage (read: offset) will pass. This is in turn inverted and substantially amplified by the opamp. C22 is placed between the input and the output of the opamp so it works as an integrator. Because there is no resistor present across C22, the amplification is maximal. The integrator will insure that there is a quick response to a changing offset. Through a second low-pass filter, that consists of R34/C23/R35, the result is passed to the input of the amplifier. The long R/C time of the filters at the input and the output of the circuit prevents that the response to changes becomes too 'nervous'.

The final offset is for a great deal determined by the properties of the used opamp. Parameters like the input bias current (Iib) and with less impact the offset voltage (Vio) are foremost responsible for a deviation of the ideal zero volts. The bias current runs through R33 and is multiplied by 1MΩ. Together with Vio, it appears as an offset voltage at the output of the amplifier. A type like the NE5534 for example, with an Iib of 500nA, will cause about 0.5V offset at the output this way. The remedy is worse than the desease in such a case. A precision opamp, or a type with FET-inputs is the best option. The specifications of most modern types are more than sufficient, Iib will be in the pA-range then. Suitable types are the OPA132 or AD8610 for example. The AD8627, a low-power SMD type, was chosen here.


A small PCB has been designed for this circuit, which can be mounted on the back of the amplifier PCB. Since only one opamp needs to be powered, the circuit can be supplied straight off the +/-75V supply voltage for the amplifier, and no extra supply transformer is needed. By means of two resistors and zenerdiodes, a symmetrical +/-12V is created for the opamp. Calculate the value of the resistors assuming a voltage drop of 63V (75V-12V) at a current equal to the supply current of the opamp, plus about 5mA for the zener diodes. For the AD8627, this calculates for 10kΩ at 0.4W power dissipation.

The connections are placed in such a way that they can be made directly on the amplifier PCB. The amp's output offset is tapped from the junction of the source resistors, just before L1/R31. The correction voltage is injected into the circuit after C1. C4 and C5 are replaced with a wire jumper. The integrator needs a few seconds of time to get the output to 0V when powering up. This is caused by the long R/C-time of R33/C22. Usually this is not a problem because the amplifier is equipped with a DC protection circuit with a power-up delay. But it is possible that the offset has not approached zero volts near enough when the loudspeaker relay is energized, and this is audible by a small pop out of the speakers. In this case, it's best to extend the power-up delay time a bit. Most of the times, this is determined by a capacitor in the delay circuit, which is slowly charged. By placing a larger capacitor or a higher charging resistor, the time can be extended.

<< back to page 1