Purifi white paper shows that a series notch can decrease breakup distortion

[capslock]

Quite interesting after cursory reading. They have an alu cone with a 5 kHz breakup and decrease distortion by adding a series notch that has about 45 Ohms impedance at 1650 Hz. Seems this higher driver impedance decreases motor distortion, so it is not at all about the mode coupling into the motor and getting short-circuited by a parallel notch. However, if it works so well at one frequency, why not use current drive in the first place? Of course, that means one has to revert to voltage drive near resonance or emply motional feedback of some sort.

https://purifi-audio.com/wp-content/uploads/2022/03/220211_R05-Notchfilter.pdf

[tktran]

Yes I've read it too. I haven't had the chance to try this, but I have a variety of drivers that have resonances between 4-8KHz that I could try it with.

My question is, is this a novel technique?

It should renew from interest in all the metallic drivers commonly used by HT Guide eg. Dayton Audio RS and all those SEAS drivers favoured in the past. All have resonances between 5-8Khz (W22EX, W18, L18, W15CY/H) etc

[5th element]

Interesting.

As far as I am aware series notch filters can also be added without the need for a low pass to function. I don't think the same can be said for parallel ones.

This implies you can add series ones to active crossover implementations too.

Parallel notches are almost always used because they don't put components in the signal path and only pass current at the tuning frequency of the notch. In contrast the series filter is always passing signal.

The purifi driver has exceptionally low distortion even with the metal cone resonances present. It will be interesting to see if this method will work with drivers having only mediocre motors. In other words just how effective can this method be! It's interesting because the component sizes for high frequency notches are very small so it's not exactly costly, or space expensive, to implement.

[tktran]

To be clear, what Lars calls a series notch, and is proposing, is an inductor, capacitor, and resistor connected in parallel, but in series with the driver, AT the very last leg of the circuit ie. directly before the driver's input terminal.

I think a nomenclature may be a problem, because in some circles this is actually called a parallel notch, because the inductor, Capacitor and Resistor components are in parallel. And a series notch is the L, C and R components wired in series to ground, in parallel with the driver.

[JonMarsh]

You know, there's no substitute for measuring verification. The notch filter on the SMJ Calliope is a good example- totally cleans up the top end HD3 induced from resonance amplification as well as the frequency response. And the Q control can be a resistor in series with one of the reactive elements.

Here's example measurements for the DA25TX, raw, and in the Calliope system









This technique was employed in the NatalieP and many other earlier designs.

[5th element]

Do you have the filter schematic for that Jon so we can compare the implementation to the white paper?

[augerpro]

Maybe Jon can interpret Lars' answer to me. I take it as "no" but I don't quite understand the mechanism he's describing.

augerpro said:
I have a question: I've often used a small .2-.3 uF cap in parallel with the first inductor in a woofer circuit to notch the woofer breakup in the 5-10khz area. This should work as well? It's a series notch, so high impedance at that frequency, does it matter that it is the first filter in the circuit? Must it always be the last?

Lars response:
if the coil with the small cap across is not last then the driver does not see the high load impedance caused by the LC tank. However, you would also normally like to have a cap to ground after the L to get a 2nd order electrical filter. This is why Kim (app not author) made the separate notch coming after the first usual series L with a C to ground. Then you both get the roll off and a load impedance peak at the notch frequency.

[JonMarsh]

Do you have the filter schematic for that Jon so we can compare the implementation to the white paper?

Sure- here's the schematic for the Calliope two way -

[5th element]

Right so the series connected notch filter, that's in parallel with the driver, looks like it's tuned quite low in frequency.

The one before that, with the tiny inductor and which is very similar in nature to the one in the purifi paper, is tuned for the ultrasonic peak. This definitely reinforces their research, which is great to see, and gives additional food for thought when using metal drivers.

[JonMarsh]

Right so the series connected notch filter, that's in parallel with the driver, looks like it's tuned quite low in frequency.

The one before that, with the tiny inductor and which is very similar in nature to the one in the purifi paper, is tuned for the ultrasonic peak. This definitely reinforces their research, which is great to see, and gives additional food for thought when using metal drivers.

The LF notch filter like that to kill the initial woofer breakup has been standard procedure for me, and given that I'm using my own concept of an LR3 since WAY back when (Modula and Natalie P) it's something I'd consider tried and true; and it's just the added parts tuned to work with the second inductor. With those quasi series crossovers, I think folks were so confused by the overall topology that they didn't really notice the notch filter.

For the Tweeter, I wind up using something in the range of 25 or 50uH. Not easy to get those values, but then you can unwind larger parts, too. I like parts that are clean enough up top to not need that (like the SS 6640, or the BlieSMa T25B-6) but there aren't many in that class.

For one design, I use a 15uH there...

[capslock]

What made you turn to a parallel XO in the Callipe then?

[JonMarsh]

Right so the series connected notch filter, that's in parallel with the driver, looks like it's tuned quite low in frequency.

The one before that, with the tiny inductor and which is very similar in nature to the one in the purifi paper, is tuned for the ultrasonic peak. This definitely reinforces their research, which is great to see, and gives additional food for thought when using metal drivers.

I think you're misunderstanding that- the network in parallel with the tweeter is an FS zobel - it's purpose is to flatten the impedance curve of the tweeter, so that their isn't interaction with the crossover filter, and the desired roll off curve can be achieved.

I have used parallel elliptic filters in a crossover in order to get a controlled high order roll off but using fewer components on occasion- such as having a nominal 4th order network but getting a 6th order curve for an octave or so around the crossover point. An example is the high pass tweeter crossover for the old M8 project.





But that was a function of C2041 and it's interaction with the shunt inductor L2041. There is also an LCR network in parallel with the tweeter, but that is also an FS zobel network.

The M8 PDF explains it fairly completely. That used to be available as a linked download, but the functionality on iCloud changed. Due to all the graphics, it's a large PDF, about 48 MB, but it's no problem to send via Apple Messages. But email will rarely work. Most systems still seem to be limited to 20MB or so.

[5th element]

I had assumed that the parallel connected notch was for flattening the impedance, at fs, but I wasn't entirely sure.

I mean a notch filter like that is always going to lower the impedance, at the frequency it's tuned at, but it's the frequency response shaping that we're usually after when that lowering of the impedance is worked on via the high pass/low pass.

[JonMarsh]

I had assumed that the parallel connected notch was for flattening the impedance, at fs, but I wasn't entirely sure.

I mean a notch filter like that is always going to lower the impedance, at the frequency it's tuned at, but it's the frequency response shaping that we're usually after when that lowering of the impedance is worked on via the high pass/low pass.

Yes, BUT... if you have a strongly rising impedance near the zeros of a high pass filter, it offsets the design of the filter because the filter is nominal designed for the driver nominal load impedance- if that is 4 ohms, but the Fs resonance pushes it to 15 or 25 ohms, (happens all the time) then the high pass function is quite screwed. The Fs zobel (impedance compensation network) targets creating a flat impedance through the Fs region, so that a simple "classical" filter will work as expected.

My normal crossover approach for literally decades (since the late 70's) has been to incorporate LCR and RC networks first to flatten the driver impedance over the entire operating range. Then, design the crossover. Often simplifies things greatly.

Look at this impedance plot of a TW29BN-4 - if you want to cross it over at say, 2200Hz, the Fs bump is going to be a problem.