Alright, where were we? Oh yeah, phones.

If you’re new here, the goal is to design a from-scratch, modular, hobbyist-friendly interconnect for analog telephones. It’s like a little phone company in a box that fits on your desk.

Part 5 didn’t go exactly to plan. The blue “SimpleSLIC Filter” boards needed quite a bit of bodging to work at all. I figured my cheeky cris-crossed CD4066 circuit was largely to blame - it’s a little weird looking but it will allow me to daisy-chain SLICs together. A completed call looks like this:

The cool thing about the above circuit is that it allows an arbitrary number of phones to share a single “trunk,” using a single commodity DIP chip per phone. the uncool thing about the above circuit is that it didn’t work. In the first iteration of the blue SimpleSLIC Filter board, I discovered that the picoamp-scale leakge current rating for CD4066 is… optimistic, to say the least. Horrible distortions made the line nearly unusable, but this seemed to (mostly) go away by adding some resistors between the switches.

While this board isn’t out of my depth intellectually, these sorts of problems aren’t things I’ve ever had to deal with on a hobby project. Analog is brutal, man.

I considered throwing the CD4066 approach out the window entirely, going to old-school relays and transformers. That’s the proper way to do phones, after all. But, relays are expensive and power-hungry, and transformers are really expensive and big, all substantial drawbacks given my hobbyist-friendly goals.

Meanwhile, the CD4066 costs a few cents on Digi-Key and it could be hand-soldered by a novice without any special tools. If I can make this design work, it would be a breakthrough. So, I took a gamble on a new board revision:

These boards don’t just add the 100 ohm resistors necessary to mitigate the audio distortions; I have also completely changed the board layout. I’ve split the ground plane in two, isolating the high-current path to the SimpleSLIC from the sensitive filtering circuitry. The audio traces are also moved further away from the power traces. I’ve also crammed the board full of ground vias. Since this is still a prototype and keeping with the hobbyist spirit of the project, I decided the ground vias would be better than going to a 4-layer design.

I was about to start assembling the boards, when I had to make an emergency trip back to the motherland (Chicago). This interregnum gave me chance to get a design review from a dear friend who’s actually a professional at this stuff. He had a whole bunch of suggestions for me, which I will be implementing in the next board revision:

1. Try to reduce capacitance between the op-amp outputs and ground. Op amps don’t like capacitive load.
2. Redo the decoupling caps, those are probably causing issues. I knew my 3-cap setup was less than ideal, I didn’t think about how that 100n cap across +5/-5 could be actively harmful.

A few suggestions I didn’t take:

3. Add line drivers after the op amps. The LM358 isn’t really meant to drive a switched line like I have with the CD4066. This is good advice, but I want to see how the LM358 holds up on its own. Part of the idea here is to keep cost down, and a pair of line drivers will add to the BOM cost and board real estate.
4. The split ground planes probably aren’t necessary, it would be better to just lay things out such that the high-power SLIC ground path doesn’t cross the path of the filters. This is empirically correct and I will do that in the final design, but these boards are meant to be cheap proofs of concept. A small board is a cheap board, so I’m prioritizing packaging over perfect ground paths.

He also told me “something feels wrong about this filter” but he couldn’t put a finger on why.

I headed back to San Jose confident that these yellow boards were going to be a huge improvement, and the next gen boards will be perfect.

I sat down at my desk, I opened up the schematic, and I noticed something. This is my Sallen-Key filter:

This is what a Sallen-Key filter is supposed to look like:

Spot the difference.

God damn it.

Next round of boards coming soon.

With the long break since the last post, this blog has proven its value to me in ways it hadn’t before. Thanks to my blathering on, I have a written record of what I was doing 9 months ago, what was working, and what wasn’t. I also feel at least a hint of accountability to the folks reading this blog (I’m assuming there are a few of you). Without this blog, this project probably would have died on the vine, so thank you for reading.