Prototype stereo balanced passive filter for Ozone
Posted 14th October 2013 at 06:57 AM by abraxalito
Updated 27th November 2013 at 02:42 AM by abraxalito
Updated 27th November 2013 at 02:42 AM by abraxalito
Here's my first stab at an affordable yet screened passive filter using off-the-shelf inductors. Its a 17th order quasi-Chebyshev using 8 equal valued series L and 9 shunt caps achieving slightly better than 50dB rejection at 24.1kHz.
Update - this might be the first and last filter I build with these Coilcraft inductors. (If anyone's interested, the part no. is LPS6235-335). Reason being - on trying to order more, the price got hiked more than 100%. Originally these were attractive because on Taobao there was a source selling well below the manufacturer's list, but that seems to have dried up. At the list price in USD, there are other alternatives potentially better value. However none with the same inductance (3.3mH). The key feature is the Q - the Coilcrafts have a Q around 40 at 20kHz - this is very hard to beat in such a small package. As a reference, my pot-core hand-wound inductors have a Q about 160 but they're huge in comparison, so these tiny inductors are actually pretty good. I've found some ever-so-slightly larger TDKs which have a Q almost 40 but are half the price of the original LPS6235s and I'm building a version at the moment. Pic to follow....
The revised filter has a BOM cost about $9 if NP0s are used and around $5 with X7Rs (which I intend to listen to, to see if there's any difference in SQ).
Update2 - I've just finished building a totally experimental even lower cost (and more compact) filter board using ferrite beads, so have added a pic. No listening yet - need to complete the input and output boards
Update3 - the hand-wound ferrite bead filter plus preamp sounded awesome (within a gnat's whisker of a purely passive stage) but I've abandoned that for now because on buying a second batch of beads I found they'd changed the formulation and I couldn't get such a high Q. So back to the TDK filter for the published version. The awesome sound (and I have to say I was staggered, its pretty hard for me to be staggered - this filter is using X5R caps, not NP0s ) with the beads might not have been due to the beads themselves as I built an integrated filter and preamp using twin AD605s and went a bit overboard with the decoupling. If you look carefully at the pic - to the left hand edge the two AD605s are buried under piles of 0805 ceramics soldered like a wall on top So the next step is to duplicate that analog stage but using the TDK coils to remove that as a variable.
In using the AD605 the input resistance is on-chip and well matched to the TDK version filter - in theory. That's provided you use both channels of the AD605 to give a differential output - then the input resistance is 87ohms per phase. But the DS says that's got a fairly wide tolerance, so I'm thinking to go even lower in inductance - down to 330uH so that the filter works into 27ohms. Then the necessary termination resistors will tend to swamp variations in the AD605's input impedance. TDK's truly excellent webpage (why can't all vendors give as much detail?) for their inductors shows the AC resistance of the 330uH part is nice and flat out to 20kHz which is just what we need. Watch this space......
Update4 - pic added of quick and dirty TDK filter.
Update - this might be the first and last filter I build with these Coilcraft inductors. (If anyone's interested, the part no. is LPS6235-335). Reason being - on trying to order more, the price got hiked more than 100%. Originally these were attractive because on Taobao there was a source selling well below the manufacturer's list, but that seems to have dried up. At the list price in USD, there are other alternatives potentially better value. However none with the same inductance (3.3mH). The key feature is the Q - the Coilcrafts have a Q around 40 at 20kHz - this is very hard to beat in such a small package. As a reference, my pot-core hand-wound inductors have a Q about 160 but they're huge in comparison, so these tiny inductors are actually pretty good. I've found some ever-so-slightly larger TDKs which have a Q almost 40 but are half the price of the original LPS6235s and I'm building a version at the moment. Pic to follow....
The revised filter has a BOM cost about $9 if NP0s are used and around $5 with X7Rs (which I intend to listen to, to see if there's any difference in SQ).
Update2 - I've just finished building a totally experimental even lower cost (and more compact) filter board using ferrite beads, so have added a pic. No listening yet - need to complete the input and output boards
Update3 - the hand-wound ferrite bead filter plus preamp sounded awesome (within a gnat's whisker of a purely passive stage) but I've abandoned that for now because on buying a second batch of beads I found they'd changed the formulation and I couldn't get such a high Q. So back to the TDK filter for the published version. The awesome sound (and I have to say I was staggered, its pretty hard for me to be staggered - this filter is using X5R caps, not NP0s ) with the beads might not have been due to the beads themselves as I built an integrated filter and preamp using twin AD605s and went a bit overboard with the decoupling. If you look carefully at the pic - to the left hand edge the two AD605s are buried under piles of 0805 ceramics soldered like a wall on top So the next step is to duplicate that analog stage but using the TDK coils to remove that as a variable.
In using the AD605 the input resistance is on-chip and well matched to the TDK version filter - in theory. That's provided you use both channels of the AD605 to give a differential output - then the input resistance is 87ohms per phase. But the DS says that's got a fairly wide tolerance, so I'm thinking to go even lower in inductance - down to 330uH so that the filter works into 27ohms. Then the necessary termination resistors will tend to swamp variations in the AD605's input impedance. TDK's truly excellent webpage (why can't all vendors give as much detail?) for their inductors shows the AC resistance of the 330uH part is nice and flat out to 20kHz which is just what we need. Watch this space......
Update4 - pic added of quick and dirty TDK filter.
Total Comments 17
Comments
-
Hi Abraxalito,
I guess this filter is done now, and if I want to build one for my self, all I have to do is to get 32pcs 3.3mH coils and the 18 shunt caps. The only dark area remaining for me is the value of the caps. Also I wanted to ask you ,is that absolutely necessary to have the filter this steep when is op amps around like the LM4562 with 55MHz bandwith?
Thanks, MiklosPosted 19th November 2013 at 07:04 PM by miklos -
Hi Miklos,
Thanks for your questions - very good ones.
On the 3.3mH version of the filter, the shunt caps I used were two 33nF NP0s in parallel, giving 66nF. At the input and output though only a single 33nF. So you'd need a total of 64 33nFs.
As for the steepness of the filter, it'll need a bit longer answer. Firstly this filter isn't designed to work with LM4562 although you could use it. I'd not recommend it though as in my experience its one of the most RF-sensitive chips I've come across. Meaning hard to get the RF levels at the inputs and PSU to a low enough level for optimum SQ. The chip I've been using is the AD605 - partly because it contains a very nifty volume control and I can get it much cheaper than list price but also because it has a much greater GBW than a typical opamp. The gain is about 50X and the bandwidth 40MHz giving 2GHz GBW.
From which you might deduce that the filter isn't particularly related to the GBW requirement of the following amp. Its steep for a couple of reasons - firstly because the image frequencies from a NOS DAC start at 24.1kHz and I want to reject these. Secondly because within the quasi-Chebyshev topology I've chosen, more stages means flatter passband ripple - and steepness comes along with more stages. You could if you like reduce the length of the filter - but then passband ripple would suffer. You'd get a less steep filter though.
Another reason for the steepness is that the downstream amplifier will generate IMD - the way to minimize this is to bandlimit the signal you send it as abruptly as possible.
There's one other detail to point out - the working (termination) impedance for this filter is 270R. So both source and load should present this value.Posted 20th November 2013 at 12:37 AM by abraxalito
Updated 20th November 2013 at 12:45 AM by abraxalito -
Posted 20th November 2013 at 12:41 PM by miklos -
Yeah the AD605 is expensive at the normal price, I doubt if I would have chosen it if not available here much cheaper. Without the volume control I've used and recommend AD8129 - I have checked Mouser and they have these about $4 in small quantities.
You'll notice that AD8129 isn't a normal opamp - you apply the output of the filter to pins 1 & 8 (the differential input) and take the feedback resistor to pin5. It has a GBW of 2GHz and must be used with at least 20dB of gain. If you drive the filter with two chips (say TDA1387s), one digitally inverted, they each give 1mA so the total signal you get out of the filter is 270mV peak-peak. If you want to create a normal CD-level output that's 5.6V peak-peak hence a gain of X20 or so is called for. Ideally the resistors at the inputs should be equal, so I recommend the feedback network to be 130R and 2k7. The AD8129 needs good decoupling - ceramics and 'lytic in parallel.
If you're using TDA1543 that has higher output current so the gain of the AD8129 should be reduced to the minimum X10.Posted 20th November 2013 at 11:21 PM by abraxalito
Updated 21st November 2013 at 12:10 AM by abraxalito -
I was thinking to modify an old Denon DCD-2560 player with four AD1862N or I also have a Sony DAT board with a TDA1541. The Denon is balanced, but the TDA1541 is not. It's a plenty of space in both case for your filter. If I will do the TDA, how would I half the filter? I guess the capacitor values has to change.
Thanks, MiklosPosted 22nd November 2013 at 12:09 AM by miklos
Updated 28th November 2013 at 09:27 PM by miklos -
The filter's actually four channels of single ended so the cap values don't need to change. Just going SE its much harder to get the grounding correct for optimum sound - I have tried and have abandoned SE as it didn't sound as clean.
With TDA1541 don't you have 4mA? I think that you'll have a problem with compliance into 135R load (the 3.3mH version) so if I were you I'd go for the TDK inductor version which presents about 45ohms to the DAC. Unless you have a common-base transistor at the DAC output to increase its compliance.
The TDK part is SLF7045T-102 (1mH, DCR 2.3ohm) and the caps you need are 100nFs instead of 33nFs.
Having had a quick look at the AD1862 DS it doesn't say anything about output compliance voltage so to be safe you'd best use the TDK version of the filter with that too. It has bipolar 1mA output which will give about 60mV RMS output in balanced with this filter - the AD8129 would need a gain around X33 were you to use that.Posted 22nd November 2013 at 03:23 AM by abraxalito
Updated 22nd November 2013 at 03:50 AM by abraxalito -
Posted 22nd November 2013 at 06:03 PM by miklos -
Posted 22nd November 2013 at 10:50 PM by abraxalito
Updated 22nd November 2013 at 10:56 PM by abraxalito -
Posted 26th November 2013 at 04:28 AM by miklos -
That was quick work - looking forward to hearing how it sounds!
Incidentally I've found some DAC outputs don't like driving capacitive loads much, so I've taken to including another inductor between the I/V resistor and the filter. Its value isn't too critical - for your filter something from 100uH to 1mH should work - you could just try with another 1mH.Posted 26th November 2013 at 08:21 AM by abraxalito
Updated 26th November 2013 at 08:25 AM by abraxalito -
Thanks for the advise. Here is a picture of the filter sans resistors in my photo gallery: Abraxalito's Balanced post DAC filter.jpg - My Photo Gallery
Posted 26th November 2013 at 02:33 PM by miklos -
Posted 26th November 2013 at 02:48 PM by miklos -
Wow, I'm impressed - that filter looks like a true work of art! I built a second filter yesterday but because I wanted it as short as possible, I didn't use a PCB at all - just soldered all the inductors end-to-end. Since they're about 0.3" long it came out about 2.5". I also soldered the 100nF caps in series across the inductors - they fit well in 1206 packages. To complete it, solder links between the centre points of the 100nFs across to the other phase. Perhaps I'll post a photo later.
Posted 27th November 2013 at 12:41 AM by abraxalito -
Posted 27th November 2013 at 02:40 PM by miklos -
I put up the pic this morning - its the fourth pic in this blog posting. The compact one is bottom left. Since then I have made it a bit sturdier, mounting it down on a piece of perfboard. I'll make another one soon so I can compare the two cap types (X5R, NP0).
By the way, I'm currently listening to the compact one with NP0s, very promising. Seems the main advantage comes from the decoupling to the AD605, not the filter type.Posted 27th November 2013 at 03:37 PM by abraxalito
Updated 27th November 2013 at 03:43 PM by abraxalito -
Posted 29th November 2013 at 12:17 AM by miklos -
Rolling up the filter? An intriguing idea - you mean like a strand of DNA - a double helix filter?
As regards resistor values, i reckon smaller should be better, limited by the noise of the amplification stage. AD8129 has about 4nV noise, roughly double that of AD605 so needs a filter with larger sized inductors to keep noise at bay. Yesterday I made an interesting discovery - the AD604 which has a preamp built in with a super-low noise of 0.8nV so I am tempted to get some to play with and find out how low the inductor values can go. Probably they"ll need winding with litz to keep AC losses down....Posted 29th November 2013 at 12:32 AM by abraxalito