Build -- Active Twin-T notch filter for distortion analysis

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
It's cool this thread picked up some activity.

I finally moved my notch filter from my prototype board to an enclosure. Here are a few photos. Thanks again Dick for this design. I tested it and it is working great.








tkollen, what caps did you end up using in your filter? Do you have the vendor and part numbers you would be willing to share?

Dave



Cool jobs....
 
Well, the trash picker in me found this headed out for 'e-recycling' and I grabbed it. It was a good size for the notch filter.

Here is what it looked like before. This company makes a few products using the same size enclosure (top is plastic and bottom is metal). I'm sure there are more than a few for cheap $$$ on eBay.

An externally hosted image should be here but it was not working when we last tested it.


I wish I had the ability to do enclosure kits. That is the biggest draw back I find...finding the right enclosure for an affordable price.

Dave
 
When I started building the Moore Twin-Notch Filter my goal was to get a filter good enough for a 60-80 dB notch which would be more than enough for its intended purposes. However it did soon also become a challenge to find out what possibly could be achieved by fairly simple means. Evaluating the transfer function of the Moore-filter revealed that a maximum notch depth of about 97 dB could be achieved with proposed values, provided perfect matching resistors and capacitors. Changing the feedback k-value by increasing the 215 ohm resistor to a higher value could increase the notch depth to as much as 110 dB +, however at the expense of a wider band width. In the end I felt that Dick's original k-value had this trade off about right ( It only attenuates the second harmonics with ca. 0.4 dB). I am thinking of making the feedback resistor switch selectable which would give me also a much narrower notch but also with less depth (ca. 60 dB).

I realized the importance of keeping the resistance equal between leg 1 and leg 3 and its relation to the leg-3 resistance to be 2/1. Equally important is it to get the capacitors to match in the same way. I noticed that only slight differences from this ideal had a significant impact on the notch. Using 1% film resistors I started out by equalizing, by using the trim pot, the difference resistance (Trim +Coarse +Medium pots) between leg-1 and leg-3 at 1 kHz setting. I then measured the difference at all combinations of switch positions and although it varied between switch positions it was never higher than 24 ohm. That's 0.35% of total resistance at the lowest switch settings and 0.03% at the highest settings (statistically the difference gets lower the more resistors are included. Compare that to the 5% capacitors that I started out using. The difference between the resistance of the legs has the same impact on the filter transfer function as the difference between the capacitors do (percentage wise). I found it even more important to match capacitors than to match resistors unless you want to bundle many capacitors in parallel (differences are cancelled out by using many of them). The proposition to buy many capacitors in order to select a few is expensive (since they are sold with either 5% or 10% tolerance) but I was lucky enough to have a bunch of older polystyrene capacitors in my junk bin to pick from and I did finally settle when I had 0.1% matches.

As I pointed out in my previous post I first tried to resolve the imbalance between the resistance in leg-1 vs. leg-3 by adding a third pot to leg-3. Although that improved measurements for frequencies between the 22 switch positions it did also become more cumbersome to operate. Instead I installed a dual 200 ohm 10-turn Helipot, which I found on eBay, for leg-1 and leg-3 (also suggested by RNMarsh in his post). The pot would cover any frequency between the 22 switch positions and with some overlap. In addition I added a second pot (100 ohm, wire wound Helipot) to leg-1 only and a 50 ohm resistor to leg-3 only in order to cancel out any imbalance between leg-1 and leg-3 at any frequency. By trying different types of pots in this project I noticed best results with the larger wire wound pots such as Helipot, Spectrol etc. Some of the smaller (and cheaper) Chinese pots are too grainy. For leg-2 I used a 100 ohm, 10-turn wire wound Helipot + 5 ohm trim pot.

Finally, I added an extra output IC for 40dB amplification and used a AD797 IC for high gain and low THD+Noise.

Finally a great thanks to Dick for this very useful tool.

Tom
 
Nice work TKollen,

Dick sent me an ESI bridge a year or so ago for matching the caps. I found the Panasonic stacked PP caps to have a zero null between pairs. Didn't care about exact value just the matching. The test was on a very small sample but I was very impressed with the results.
These are 3% and 5% parts.
 
Member
Joined 2004
Paid Member
Here is an update on my progress with Jens notch filter.

Phase one was matching the filter requirements to the parts I have on hand. After working with the B&K I decided I needed to lower the impedance of the filter to get lower noise. In my collection of parts is an assortment of Tektronix caps. These are spares for some scope plugin. I have a handful of them. They seem to be hermetic polystyrene 1% caps. They all measure withing .5% of the marked value (after 30? years). I have .1uF X 3 so I started there. Then I used this tool: Twin-T Notch Filter Design Tool to calculate the resistor values. Its limited to standard values so I started there and put the whole circuit into LT Spice (asc file attached).

I tweaked it to get the following properties:
1) 40 dB attenuation at the center frequency (1KHz)
2) As wide a “flat” band at the center frequency as I can get, about .5% or 5 Hz at 1 KHz with about +/- 3 dB in the band
3) 40 dB gain at the harmonics, the net effect is boosting the harmonics by about 40 dB and still having the fundamental. This allows using a conventional audio analyzer with a 100 dB floor for distortion to have a -140 dB distortion floor (on paper).

Assembling the board SMT parts is not easy for me at this stage, even with the microscope and tweezers I find these parts frustrating, Especially after the bigger parts are installed. In any case I got the active parts assembled on the PCB. I used an LME49990 for the main amp for its low noise and an LME49710 for the Q multiplier since I had it.

Other experiment- can I run this on battery power with no center tap? The answer is yes. I used 2 10K resistors, one from each rail to ground. The applied power from rail to rail and its all behaving quite well. There should be no DC at the output. The output can be AC coupled with no down side in fact.

I then made up the tuning network with all the parts exactly to value (you can see the strings of resistors with trimmers to get them on value in the attached photo) and tested the network. It all works as planned. Its really sensitive to external noise so it will need to be built on a tight layout on perfboard and all in a can to be useful. But the important part is that it all works. I verified the distortion measurement using a function generator I have. It has about .3 % THD. Through the filter it measures 30% or 40 dB higher.

With some focus this week I should be able to measure DavidA's oscillator by the end of the week down close to its noise floor. It already outperforms the Shibasoku 725 so I'll be using the 725 with this notch to get deeper into the floor.
 

Attachments

  • Twin Tee.PNG
    Twin Tee.PNG
    40.2 KB · Views: 396
  • Optimized 40 dB 5 Hz wide 1 KHzTwin T.asc
    3.1 KB · Views: 72
  • IMG_20140803_121635.jpg
    IMG_20140803_121635.jpg
    643 KB · Views: 272
Member
Joined 2009
Paid Member
You can probably get a lower noise level by using the LME49710 for the main amp and the LME49990 for the Q amplifier. In my simulation (in Tina TI) I get a noise over a 20.5 kHz BW, which is 1.25 dB lower. If you use the original OPA209 for U1 it is even lower, with a reduction of 1.75 dB. You may not need a lower noise floor, but it should be possible to do.

A passive version, with just the 40 dB gain (OPA209), of course has a much lower noise, but it is more difficult to work with, since the level of the harmonics need to be corrected in this case.

You may not need a perfboard. Perhaps you can mount some of the large capacitors on the back side of the PCB?
I agree that the circuit is sensitive to external noise. But there seems to be room for improvement in your test set-up ;)
 
Member
Joined 2004
Paid Member
Andrew- EssB
Its actually a balancing act between R10 and the ratio of R7 and R8. I tried a lot of small tweaks each way to get it where it is. Changing R10 affects R1, R2 and R5 altering the tuning a little so its back and forth to get it working. All the parts need to be very precise to get it to work. I'm sure there are other solutions that will work as well.

Jens
I'll try the OPA209 as soon as some samples get here. The odd part is that the OPA 209 has about 6 dB higher noise at 2.2 nV/rtHz and simulates with lower noise. It must be the current noise that is dominating. The OPA209 has about 1/4 the current noise.

I rescaled the network using 1 uF caps dropping the resistors and that lowered the noise some. Using the lower noise opamp for the q multiplier made a bigger difference. I think I understand but it was not obvious at first. The 1 uF caps I have are really large (the three would be the same footprint as the whole PCB) so I'll continue with the external network for now. It will also make swapping networks for different frequencies etc. easier.
 
Member
Joined 2009
Paid Member
Yes, the nice thing about the OPA209 is the relatively low current noise. With an impedance of a few kohm this gives a lower noise than many "super low noise" op-amps.

If you take the voltage noise and divide it by the current noise you will get a pretty good idea of where the optimum noise performance of an op-amp will be.
 
Here is an update on my progress with Jens notch filter.

Phase one was matching the filter requirements to the parts I have on hand. After working with the B&K I decided I needed to lower the impedance of the filter to get lower noise. In my collection of parts is an assortment of Tektronix caps. These are spares for some scope plugin. I have a handful of them. They seem to be hermetic polystyrene 1% caps. They all measure withing .5% of the marked value (after 30? years). I have .1uF X 3 so I started there. Then I used this tool: Twin-T Notch Filter Design Tool to calculate the resistor values. Its limited to standard values so I started there and put the whole circuit into LT Spice (asc file attached).

I tweaked it to get the following properties:
1) 40 dB attenuation at the center frequency (1KHz)
2) As wide a “flat” band at the center frequency as I can get, about .5% or 5 Hz at 1 KHz with about +/- 3 dB in the band
3) 40 dB gain at the harmonics, the net effect is boosting the harmonics by about 40 dB and still having the fundamental. This allows using a conventional audio analyzer with a 100 dB floor for distortion to have a -140 dB distortion floor (on paper).

Assembling the board SMT parts is not easy for me at this stage, even with the microscope and tweezers I find these parts frustrating, Especially after the bigger parts are installed. In any case I got the active parts assembled on the PCB. I used an LME49990 for the main amp for its low noise and an LME49710 for the Q multiplier since I had it.

Other experiment- can I run this on battery power with no center tap? The answer is yes. I used 2 10K resistors, one from each rail to ground. The applied power from rail to rail and its all behaving quite well. There should be no DC at the output. The output can be AC coupled with no down side in fact.

I then made up the tuning network with all the parts exactly to value (you can see the strings of resistors with trimmers to get them on value in the attached photo) and tested the network. It all works as planned. Its really sensitive to external noise so it will need to be built on a tight layout on perfboard and all in a can to be useful. But the important part is that it all works. I verified the distortion measurement using a function generator I have. It has about .3 % THD. Through the filter it measures 30% or 40 dB higher.

With some focus this week I should be able to measure DavidA's oscillator by the end of the week down close to its noise floor. It already outperforms the Shibasoku 725 so I'll be using the 725 with this notch to get deeper into the floor.

Those Big silver things are what? who made those caps?
I guess their tolerance must be tight eh? >5% How they sound?
 
Member
Joined 2004
Paid Member
I have no idea who made the caps. My best guess would be Component Research but ?? I don't want to know how they sound. Since they are unique its not useful.
They measure better than .5% after 25+ years.

I looked into using the 1 uF caps but at 1 KHz the input Z becomes about 150 Ohms so not useful.
 
Member
Joined 2004
Paid Member
I'm not sure what material MKT is. The issues would be:
1) Linearity. If the cap distortion exceeds what you are looking for its a problem.
2) Stability. Both temperature and time can be a problem.

Since its so sensitive to matching its important that all the parts match in material, tempco and probably manufacturing. That way they will track better as they age. High precision parts go through a number of steps to pre-age them so they are stable. I'm not sure the .1% metal film resistors get that treatment and would be within .1% after 5 years or several temperature cycles. Caps are more of an unknown. If they have a few years on them its a good start. You do need a very good high resolution bridge to really test and sort them. I will be padding the caps to get to the ideal value (they are all a little small which is good). I will probably have trimmers on all 4 resistor legs. I would on the Q multiplier but its SMT so not for now.
 
Member
Joined 2009
Paid Member
MKT (polyester) is not the best choice.

For best performance use polystyrene. Other solutions could be NP0 or polypropylene. I use polystyrene for the oscillator and the notch filter.

Polystyrene capacitors are a bit difficult to find, but it is possible on ebay, alibaba and other places.
 
MKT (polyester) is not the best choice.

For best performance use polystyrene. Other solutions could be NP0 or polypropylene. I use polystyrene for the oscillator and the notch filter.

Polystyrene capacitors are a bit difficult to find, but it is possible on ebay, alibaba and other places.

Where is there a good supply of polystyrene caps. I checked at Mouser and it seems these are discontinued. There are a few left here and there but I think production of them is thing of the past. They just can't withstand heat without destruction and because of this have lost their popularity.
 
Here is an alternative to using positive feedback with a TT notch.
This is from the Shibasoku patent. Instead of using positive FB a post equalizer is used to multiply the Q. This can be done with passive circuitry avoiding the noise associated with PFB. It adds to the tuning complexity but the trade off of noise might be worth it.

See 12a and 12b
 

Attachments

  • Shibasoku.JPG
    Shibasoku.JPG
    58.5 KB · Views: 268
  • US4417310.pdf
    64 KB · Views: 91
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.