As promised, I was going to share some FR charts with some text and an explanation of the low-end peaking of a transformer and how I "tamed" it. The attached charts were taken from two different KM84-style circuits with an ASTDS T-8 transformer. The relevant part of the KM84 circuit is depicted in the schematic below. Please ignore the component values of C2, C4, C9, and R13 for now: they are not set in stone and are the variables we want to play with to shape the FR and THD to our liking. C4a is an alternative for C4 and only one is placed. There is also a third chart that was taken by just using the REW tone generator and directly coupling it into the transformer through a 6k8 resistor and C4. The 6k8 resistor simulates the KM84 JFET stage output impedance.
The attached zip file contains the three REW .mdat files. File "C4.mdat" shows how the value of C4 and a significant output impedance of a driving stage affect the LF response of the transformer. The file "KM84+ C2.mdat" contains charts showing the effects of different values of the feedback capacitor C2. The charts from "KM84++ C4-C9-R13.mdat", show the effects of different coupling capacitors (C4 in series with C9) and the effect of resistor R13, which is in parallel with C9.
You can open the .mdat files in REW and view the plots. First, select the "All SPL" view to see all the measurements. They should be aligned at a center frequency of 1kHz to clearly see the effects on the FR, independent from the different gains resulting from the component variations. If they are not aligned, then select the SPL scale in the top-left corner of the All SPL view. Then, in the top-right corner, select Actions > Align SPL... to align all graphs. On the Distortions view, you examine the effects of the component variations on THD. In the case of the C4.mdat file, the distortion data is not very relevant, because the JFET distortion is not included.
I will briefly go through the data and explain what to watch to prevent a too-long post. A more elaborate discussion will be published on my website in due time. There's much more to be told...
Let's start with C4, ignoring C9 and R13. Because the JFET impedance converter in a standard KM84 circuit has a significant output impedance, I measured the effect of C4 combined with a 6k8 series resistor. Rod Elliot suggests that you should use the largest series capacitor possible if you want the flattest possible frequency characteristic (https://sound-au.com/articles/audio-xfmrs.htm#s3). That was my first instinct too, but I have changed my opinion. The example Rod describes is valid when the driving stage has a low output impedance, which is definitely not the case in KM84-style circuits! So when you compare the 4.7uF and 47uF plots, they are almost identical. With a large value for C4, the low-end cut-off is defined by the RL High-Pass response, defined by the output impedance of the JFET stage and the transformer induction. There is no resonant peak. Only the third plot with C4=0.47uF shows some peaking, which may actually be desired if you want some bass lift from an otherwise thin-sounding capsule. In the KM84, C4 is 1uF, which looks like a good choice for a flat frequency response while at the same time preventing transformer saturation from sub-sonic rumble, which would pass C4 if it is bigger than 1uF. My conclusion is that it's quite obvious that bigger is not always better!
As explained above, the output impedance of the JFET stage strongly determines the low and roll-off. The output impedance is mainly determined by the feedback capacitor C2, which also determines the gain. This is quite obvious from the "KM84+ C2.mdat" graphs. I had a zobel network attached to the secondary of the transformer (sorry, forgot the RC values), and you can also see the effects of C2 on the high-end roll-off when the transformer is capacitively loaded. You can also see that peaking is almost non-existent, and only with the largest value of C2, hence the lowest output impedance, do we see some peaking. I'm not sure what C4 value I had with this measurement. Thought it was 4.7uF, so low-end roll-off is determined by the output impedance of the JFET and not by the LC High-Pass filter formed by C4 and the transformer inductance.
Finally, and now it's getting interesting, a KM84 circuit where I increased the bias current to 2mA for lower noise. This circuit has a much lower output impedance than the classic KM84, so we can expect more peaking. When you open the "KM84++ C4-C9-R13.mdat" file, this can be seen in graphs 1, 2, and 5. If we want to get rid of this ringing, you'll have to add some dissipative element somewhere in the circuit. I chose to add another capacitor C9 and a parallel resistor R13. With C4 >> C9, C9 determines the cut-off frequency and C4 only serves as a DC-blocking capacitor. R13 dampens the resonant circuit and prevents the peaking. See chart 3
Give C9 a very small value, combined with a high value for R13, and you can use it as a low-cut filter (graph 4 without resistor => some peaking, graph 6 with high resistor value). This is a steep ~12dB/oct HPF. Suitable to reduce LF rumble and handling noise. A less steep roll-off of ~6dB/oct can be achieved by reducing the value of R13 (graph 5). Such an HPF might be more useful if you want to reduce the proximity effect.
As a reference, I added the FR plots of the CM-63 impedance converter with and without the HPF engaged. Plots 8 and 7 respectively. With the values chosen for C4, C9, and R13 in the KM84-style circuit, the FR plots are quite comparable. So these values I will use, and not the ones depicted in the schematic.
Has anyone ever seen a mic circuit with the additional R and C to eliminate the resonant peaking? Or is this a solution to a non-problem, and some peaking is sufficiently masked by the natural capsule roll-off?
Cheers, Jan
