The accuracy of the measurements with this instrument is reduced if frequency components above 200 kHz are present at significant levels. This is the case with our amplifier, particularly at low signal levels. We immediately went for large artillery, namely a 9th-order elliptical filter. During the design of the filter we made use of normalized tables. In the end it became a filter with identical termination impedances, which unfortunately means an attenuation of two times within the pass band. When converting to realistic values we selected pure E12-series values for C1(+C2).
All capacitors are arranged as two in parallel in order to closely approximate the calculated value. This applies to the resistors as well. With the inductors there is no way to avoid ‘funny’ values and series or parallel connections don’t make much sense because to achieve a certain quality, standard coils are not appropriate. So we had to think of a solution ourselves. The input and output impedance are theoretically 1.060 k and are approximated quite well with components in parallel (1.05996 k). By making use of a voltage divider it becomes possible for R3 to handle a higher voltage (otherwise note the dissipation of R1!).
Any voltage divider needs to have an output impedance of 1.06 k (R1//R2//R3). In the last section, the parasitic capacitance of the connecting cable and input impedance of the analyser has been taken into account. Trimmer C19 can be used to compensate the attached capacitance and R5 can be omitted if the input impedance is about 100 k. A deviation of about 50 pF makes little difference to the amplitude characteristic in the pass-band. The advantage of an elliptical filter compared to, for example a Chebyshev filter, is to trade off a limited attenuation in the stop- band to a much steeper transition from pass- to stop-band.
It suffices to mention that the curve from 180 kHz to 200 kHz falls by more than 60 dB, quite steep and certainly not bad for a passive filter! In practice the attenuation in the stop-band at –63 dB was a little lower than the theoretical value of 60.2 dB, which was the design value. Frequency characteristic A shows mainly the stop-band and the characteristic behaviour of an elliptical filter can be clearly seen. Frequency characteristic B shows an enlarged version of the ripple in the pass-band, which also shows the phase behaviour of the filter (scale on the right). At 20 kHz the attenuation is only 0.1 dB and the phase shift only -30°.
The first dip of only -0.263 dB occurs at about 46 kHz and the attenuation at 100 kHz is only 0.276 dB. Above that, the non-ideal behaviour of the components becomes noticeable and the curve starts to drop a little too soon, but the characteristic elliptical behaviour is still clearly visible at 180 kHz. The filter proved to be quite useful in filtering the PWM signal and analyse the LF- amplitude. The only disadvantage is the increasing distortion at 20 kHz (from 0.5 V input signal) so that good THD+N measurements can only be done at 1 kHz. This can be seen clearly in Graph C.
With 1 W into 8 (2.828 V) the distortion at 1 kHz is less than 0.001%, but at 20 kHz the distortion is 20 times larger. In this measurement the maximum input signal was 13.33 V (maximum from the analyser). For those who love to experiment and wind inductors, we have also designed a PCB. A low permeability core material (TN23/14/7-4C65) was selected for the inductors, so that saturation and material properties are less of a problem. Unfortunately this results in a higher number of turns, but also means that the inductor value can be made more accurate.
A larger core may have resulted in a lower distortion, but it would have been harder to obtain an accurate value. Toroids were selected to minimise mutual coupling — that this was successful is shown in Graph A. It is easiest when winding the cores to calculate the amount of wire required beforehand and then add 10 or more centimetres. You have to wind tightly and put the turns close together to prevent the second layer dropping in between the first layer. This applies to the inside of the ring core. When using 0.5 mm enamelled wire the second layer turns easily fit between the first layer turns.
The PCB has been designed such that connections can be made in several places (3 inside a quarter circle). The capacitors are 1% silvered mica types with an operating voltage of 500 V. That way even extreme voltage peaks will not cause any harm. There is also room for 1% tolerance ‘Styroflex’ (polystyrene) capacitors from Siemens (that are not made any more), which we have used in the past. Other manufacturers also use this shape.
Resistors:
- R1,R4 = 1.7k
- R2,R5 = 113k
- R3 = not fitted *
- C1,C14 = not fitted *
- C2,C5,C11,C13 = 1nF 500V
- C3,C8,C12 = 120pF 500V
- C4 = 6pF8 500V
- C6,C15 = 270pF 500V
- C7,C9 = 680pF 500V
- C10,C18 = 180pF 500V
- C16 = 220pF 500V
- C17 = 470pF 500V
- C19 = 100pF trimmer
- L1 = 1mH15, 115 turns of 0.5mm dia.
- L2 = 689µH, 89 turns of 0.5mm dia.
- L3 = 557µH, 80 turns of 0.5mm dia.
- L4 = 802µH, 96 turns of 0.5mm dia.
- K1,K2 = cinch socket
- * see text
- R1//R2 = 1.060 k
- R4//R5 = 1.060 k
- C1+C2 = 1.000 nF
- C3+C4 = 128.0 pF
- C5+C6 = 1.277 nF
- C7+C8 = 809.0 nF
- C9+C10 = 860.4 nF
- C11+C12 = 1.125 nF
- C13+C14 = 996.8 pF
- C15+C16 = 492.7 nF
- C17+C18+C19 = 742.4 pF
- L1 = 1.148 mH
- L2 = 693.3 µH
- L3 = 556.4 µH
- L4 = 809.6 µH
No comments:
Post a Comment