EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: taydin on October 03, 2017, 08:04:31 pm
-
Hi,
I set out to build a high quality low pass filter with the following characteristics: completely flat between 0 - 50 KHz, then drops reasonably sharp so that attentuation is around 60 dB at 200 KHz. This is going to be used to filter the output of a class D amplifier before presenting it to an Audio Precision analyzer. By the way, AP sells such a filter (AUX-0025), but it probably costs an arm and a leg, so I will try to build one myself if I can.
I am going to build this as an L/C bessel filter with 5 poles and have calculated the following values:
L1 = 16145 uH
L2 = 35364 uH
C1 = 55.48 pF
C2 = 255.92 pF
C3 = 718.81 pF
My question is, to get the best performance (above attenuation curve, low noise, low distortion etc) what type of capacitor and inductor should I be using?
-
What magnitude source/load impedance are we talking about here? The component values look suspiciously low.
Also, for maximum amplitude flatness, Butterworth filters are normally the choice.
Personally, I like Bessel for phase linearity and impulse response, but amplitude flatness and rolloff are not its strengths.
-
The class D output impedance is in the order of 3 Ohms. The AP offers three input impedance choices: 300 Ohm, 600 Ohm, 100 KOhm
The other filter types resulted in higher values for the components (especially inductor). That's why I picked the bessel. When calculating, I had used an arbitrary 10 KOhm I/O impedance for the calculations.
-
Just recalculated for 300 OHm I/O impedance:
http://www.wa4dsy.net/cgi-bin/lc_filter3?FilterResponse=Lowpass&poles=5&cutoff=50&funits=KHZ&Z=300 (http://www.wa4dsy.net/cgi-bin/lc_filter3?FilterResponse=Lowpass&poles=5&cutoff=50&funits=KHZ&Z=300)
Much lower values for the inductor ... With these values, it might be possible to use air wound inductors.
-
I think you have to go back to the drawing board here.
The AUX-0025 specifies a preferred load impedance of 200 kohms, 100 kohms is also OK.
Source impedance should be 2 ohms or less (I'm sure you'll get away with 3 ohms, no problem).
But apart from filtering, there's obviously major impedance transformation going on here, which makes your filter design void.
Impedance transformation combined with filtering is absolutely possible (and well known) with passive LC circuits, but needs somewhat more advanced mathematics/synthesis than your filter design package seems to have.
-
http://cranialcrumbs.blogspot.com/2015/05/audio-precisions-class-d-filter.html?q=audio+precision (http://cranialcrumbs.blogspot.com/2015/05/audio-precisions-class-d-filter.html?q=audio+precision)
-
Using the source for damping the filter is probably not a good idea, because of losses (high transient currents).
Bessel has the slowest rolloff, and choosing based on component values doesn't seem consistent with any other specs as presented.
No filter can be perfectly flat within a finite range, so I'm not sure what you're going for here. Would you actually be okay with a modest say 0.5dB ripple?
Tim
-
Bessel has the slowest rolloff, and choosing based on component values doesn't seem consistent with any other specs as presented.
Well, it becomes important, because very high inductor values are much harder to get while also keeping the Q high. For example, 1 mH is easy to implement air wound, while 35 mH is harder and might require a core.
No filter can be perfectly flat within a finite range, so I'm not sure what you're going for here. Would you actually be okay with a modest say 0.5dB ripple?
The main measurement I want to do is THD, so I guess it's ok if the filter isn't very flat. low noise and distortion are much more important than measuring the signal level.
-
http://cranialcrumbs.blogspot.com/2015/05/audio-precisions-class-d-filter.html?q=audio+precision (http://cranialcrumbs.blogspot.com/2015/05/audio-precisions-class-d-filter.html?q=audio+precision)
Wow excellent find! Thanks
-
http://cranialcrumbs.blogspot.com/2015/05/audio-precisions-class-d-filter.html?q=audio+precision (http://cranialcrumbs.blogspot.com/2015/05/audio-precisions-class-d-filter.html?q=audio+precision)
Wow excellent find! Thanks
That was a nice read, seeing all the challenges involved in making something like this work. And also seeing that I was way over my head trying to do this myself :o
Guess I will have to inquire about how much this unit costs
-
Well, it becomes important, because very high inductor values are much harder to get while also keeping the Q high. For example, 1 mH is easy to implement air wound, while 35 mH is harder and might require a core.
No filter can be perfectly flat within a finite range, so I'm not sure what you're going for here. Would you actually be okay with a modest say 0.5dB ripple?
The main measurement I want to do is THD, so I guess it's ok if the filter isn't very flat. low noise and distortion are much more important than measuring the signal level.
Oh -- great! Then low distortion cores (bobbin and rod ferrite, powdered iron toroid) will be fine, and Q factor won't be a big deal in the first place as filters only need high Q (over 5 or 10) for much sharper cutoff.
Lower Q parts limit the sharpness at cutoff, and introduce loss in the transition band. But if you don't need terrific flatness in the, say, 20-100kHz range, who cares?
Indeed, you usually want loss in the filter, to help isolate its response from the load, which can be highly variable. A speaker is very inductive at high frequencies, so does not act to terminate the filter; conversely, a piezo tweeter is very capacitive, and can do nasty things to a poorly compensated amplifier, or to the response of a filter! This is accomplished by adding R+C dampers in parallel with capacitors, and R||L dampers in series with inductors.
The goal is to have the filter's nominal impedance somewhat higher than the amplifier's ideal load, so that reactive current into the filter does not have a big impact on SOA and losses (it consumes only some of the amplifier's peak current capability). Then, the load impedance is isolated from the filter, using a series R||L damper. Finally, the filter is terminated with an R+C, where C is several times the filter's total capacitance.
Also, you need an LCLC filter, so an L is facing the switching inverter. (Assuming it's a conventional constant-voltage inverter. It is possible to build it so that a C-input filter is required, but that takes more effort, so it's almost never done.)
For an 8 ohm amplifier, you might choose a 16 ohm filter. The filter prototype must be a one-side-shorted, singly terminated type. I don't know of any online calculators that offer this type, but it's tabulated in Zverev and others. The mean inductance will be on the order of (16 ohm) / (2*pi*(50kHz)) = 51uH, with individual inductors taking values above and below this value as needed (the mean is actually the geometric mean).
Another tip: you can always use a signal generator to test the response of the filter, and calibrate out its frequency response. This is easiest with spectrum analyzers, but I don't know if your intended test has such capability.
Tim