Changing the filter to remove harshness

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I'm looking for some help with my filter. Both units are ScanSpeak (18W8546 en D2904/9900). I really like the lows en low-mids, very detailed. The higer-mids and highs are a little on the forward/harsh-ish side of neutral. Most apparent with trumpets and female voices. Can someone advise me in a direction to change the filtering to remove this problem?

Thanks, Ralph
 

Attachments

  • filter.jpg
    filter.jpg
    21.1 KB · Views: 629
Jan, thanks great info.

Removing harshness is done by implementing an impedance compensation network. A capacitor-resistor network in parallel with your tweeter.

François, does this mean that I add a CR network to the existing 18R and use this formulas to calculate C and R:

R = Re + (Qes * Re / Qms)
C = 1 / (2 * pi * Fs * Qes * Re)

L = Qes * Rs / (2 * pi * Fs)


ralph
 
From my experience, the harshness (and forwardness) comes from peaking in some frequency range that spans approximately 2 octaves (500Hz - 2kHz, 1kHz - 4kHz, for example). If you can source a graphic equalizer (or better a parametric one), you may try reducing a few DB in certain frequency regions to see if harshness is gone. Then, you can add certain impedance equalization (R-L-C) to certain parts of your crossover to obtain the result without using EQ.

Zobel network (R-C) will gradually drop SPL as frequency increase. While it could solve the harshness, it sometimes excessively dull the high frequency sound. Personally, I prefer maintaining flat SPL at high frequency (5kHz up) as I see that it's critical to the airiness of the sound.
 
That's a Scan-Speak refference monitor, right? If you see the impedance response of this system, there is a bump at 2kHz. That's where the harshness come from, the frequencies of female vocals. Although there is a bump, this filter is phase linier, so I will prefer to improve the performance of the amplifier or source than changing the filter or adding a compensation notch filter (RLC, not zobel/RC). But adding a notch filter (in the input of the filter) seems to be the most promising and easier approach (I guess a pair of good notch filter around 2kHz will cost almost $50, cheap? Then go ahead ;))
 
Jay,

so I will prefer to improve the performance of the amplifier or source than changing the filter or adding a compensation notch filter (RLC, not zobel/RC).

That's a good one ;) CDP and amplifier are more or less fixed and should do the trick.

But adding a notch filter (in the input of the filter) seems to be the most promising and easier approach

Do I need measuring equipment for this? Don't suppose to go to the shop and ask for a 2K notch filter, am I right? ;)
 
François said:
Removing harshness is done by implementing an impedance compensation network. A capacitor-resistor network in parallel with your tweeter. Ideally you have to measure the driver impedance response.

could u please draw this ckt?

Jay said:
That's a Scan-Speak refference monitor, right? If you see the impedance response of this system, there is a bump at 2kHz. But adding a notch filter (in the input of the filter)

what sort of notch filter series or parallel.

i looked at the the XO and noticed that the XO is effectively a 3.9uf cap and a 0.82mh inductor. while the tweeter starts getting rolled of early, the 0.82 mh inductor means that the roll off is not fast enough. try reducing the 0.82mh to 0.7, or 0.6 or 0.5mh. see if that helps.

BTW should apply any solution to the 500Hz impedance peak or jus tleave this peak alone?
 
Originally posted by Jay
That's a Scan-Speak refference monitor, right? If you see the impedance response of this system, there is a bump at 2kHz. That's where the harshness come from, the frequencies of female vocals.

Jay has a point, that's probably where the problem is. It is one of the characteristics of Kevlar cone, especially when they try to stretch the upper frq - cone excitation at breakup. This results in that 'harshness' in female sibilance. Somewhat similar to crossing a tweeter too low, thereby inducing resonance.

Originally posted by namui
Zobel network (R-C) will gradually drop SPL as frequency increase. While it could solve the harshness, it sometimes excessively dull the high frequency sound.

Worth giving that a try.

Cheers
 
That's a good one ;) CDP and amplifier are more or less fixed and should do the trick.

Tube amp! And a good one! Oh la la... Why not use better speaker? :D Well, I have a policy to go with tubes only if I have $$$ in my pocket, or I will have too many problems with my system.

Do I need measuring equipment for this? Don't suppose to go to the shop and ask for a 2K notch filter, am I right? ;)

Damm, you are correct again :D


what sort of notch filter series or parallel.

Theoritically, except for the effect on system impedance, parallel or series are the same. But it will be dificult (if not impossible) to implement a parallel one here. Anyway, I have never seen a parallel notch except in my system, and I have never built a series one either.
 
It was may fault not asking first what amp is used to drive this speaker.
The combination of bumpy impedance and the intrinsically low damping factor of SETs will give a hump in the resulting amplitude response. I assume the speaker would be ruler-flat on an SS amp.
I also don't think that it is a driver anomaly since these Scan-Speak Paper/Kevlar divers belong to the most well-behaved ones available.

It is the speaker's impedance that has to be made flat !! This can't be done absolutely exact with one series network alone due to the asymmetric nature of the impedance bump, but a significant improvement is still possible.

Regards

Charles
 
Did you measure the impedance-graph by yourself or is it from the kit-supplier ?
The reason why I am asking is the fact that we have to know the value on top of the impedance peak (and the shape of the cut off part as well). This is because we have to find the apropriate "Q" value for the series resonant network.

BTW: Almost all amps profit from a flat impedance plot, not only SET. I.e. ironing out impedance irregularities makes a speaker more "universal".

Regards

Charles
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.