Analog video input isolation

[zerosignal]

Hi all,
Great show and community on these forums.

I've been working on an RGB SCART AV switch for my growing range of retro consoles during the long lockdown months. It's an 8 input, dual output device controlled via an ATSAMD21J18A MCU. Video amplification/splitting is performed via a THS7374 video amplifer configured for dual output as per the datasheet. Inputs are switched by the MCU using one SN74CBT3244 octal FET bus switch per input. I'm routing RGB video, L+R audio plus the function and switching signals from the inputs (my SCART cables carry only a max 5V so it is safe currently in my setup). Additionally, and I think crucially to my current issue, I am connecting all GNDs of the SCART input to a single largely uninterrupted ground plane within my 4 layer PCB.

The device seems to work fairly well so far with a bunch of RGB consoles tested using shielded SCART cables from RGC in the UK. However, I notice that when more than one console is powered on at a time, the image on the selected one catches some interference from others that are powered on. I think I'm either seeing some leakage onto the bus from the inactive bus switches themselves, but more likely the current grounding solution introduces ground loops because of the lack of isolation between the inputs.

Does anyone have any advice on a) confirming which is the most likely scenario and b) how to properly isolate the inputs? Given they are analog video/audio signals (and a lot of them) optocouplers dont feel appropriate without some form of ADC/DAC conversion (which feels counter-intuitive to me, not to mention expensive). I've looked into isolation amplifiers and again they don't seem the right solution to me in this case. I'm wondering whether a more appropriate solution would be to move towards a star grounding arrangement on the inputs.

I've made a decent effort on the PCB layout to carefully route the signals and minimise crossover etc, and as I say it seems to introduce no noticeable degradation in video quality when added into the chain with just a single console powered on. I'm interested in fixing this if possible, as the next stage of the project is to implement auto-switching functionality when an additional input receives a valid video signal.

Attached a schematic of one of the 8 inputs (they are all the same) for reference. The TVS diodes on the schematic are currently unpopulated on the single prototype of the switch that I've built. Any help would be greatly appreciated!

[ogden]

Proper routing of hi-frequency signals is not easy job. It still could be crosstalk due to poor return ground for hi-freq signals. Do you treat SCART signals as RF signals requiring 75 Ohm transmission line - coax or microstrip? Before you continue with project, you may want to watch this video from Altium. Any chance you show photos of device/PCB?

[edit] Why don't you just buy SCART input multiplier / switch box?

[zerosignal]

Thanks for the tip, I will watch the video. Attached a screenshot of part of the PCB, SCART inputs on the left of the image with the two outputs on the right.

Fair point about buying one, most of the ones currently available are of fairly low quality on the cheaper end and are becoming harder to get hold of. There are some that are made specially for retro gaming, which in hindsight would have been cheaper to buy at this point, but it was just a hobby project of mine that I started because I wanted to learn some bare metal ARM development.

[ogden]

First thing I notice - bus with stubs. Every stub line connected to bus is stray capacitor+inductor combination and it will impact frequency transmission characteristics of the bus. I would put switches as close to the bus as possible, rather right on top of the bus. Also make sure that below every (video) signal line there's ground plane for return currents. Make sure your microstrips have 75 Ohms impedance. Is this 4-layer board with solid ground plane? Like this

[zerosignal]

I understand what you are saying about the optimum placement of the switches, will look to address that in the next revision. So, in summary, it would be preferable to have longer traces between the SCART connector and the switch, than the switch and the bus itself? The board is constructed as in your link, with the solid ground plane being adjacent to the (green) signal layer.

Really grateful for the advice so far! I'm still wondering why there is noticeable difference when another console is switched on at the same time - do you think the ground loop theory is not the cause here?

Watching the video now!

[Renate]

Yeah, it takes a bit of work to make a clean analog video router.
One of my favorite routers from years gone by was by "International Nuclear".
I've searched but can't find any photos.

On a metal chassis, one side had all the source feeds (from preamps) running along a busbar.
On the other side, perpendicular you had all the destinations (to drivers) running along a busbar.
At the junction you had a reed relay stuffed halfway into a hole through the chassis.
Each side had big grounded fins separating all the bus bars.
So the only crosstalk that there was was the tiny amount from an open reed relay.

Mmm, I think the long stubs on the bus will add to ringing, but I think the crosstalk could be mostly? crosspoints?
I haven't looked at the spec. Send in 5 MHz and see how it bleeds through.
Of course more ground is always good.

[zerosignal]

Do you mean crosspoints where the switches connect in to the bus where the red traces cross the green in the picture?

[ogden]

So, in summary, it would be preferable to have longer traces between the SCART connector and the switch, than the switch and the bus itself?

Yes. BTW - do you use RGB or composite?

I'm still wondering why there is noticeable difference when another console is switched on at the same time - do you think the ground loop theory is not the cause here?

Hard to tell from where I am now. Poor isolation of switch IC's could be problem. If there are any - check for alternatives, compare datasheets. You need switches with best "off" isolation possible. While testing - check various main/interferer signal combinations, better in every possible two port combination, write test report table. If there are differences that depends on ports used - it may give you valuable hint about possible source of the problem. Also try to analyze interference as best as you can - does all the signals of SCART bleed through? If so then poor isolation of switches most likely to blame. If only one signal of whole SCART - then who knows, could be some routing or ground problem too.

[ogden]

To be honest I would do tree like structure rather than bus. To build 4:1 switch I would use three 2:1 switches, like MAX9655 which is purpose-built SCART switches IC.

[edit] Just noticed that you use SN74CBT3244. That is not analog switch. Video signals are ground-referenced AC signals. Your switch is designated for logic signals, between 0V and VCC. Easiest way to overcome this - introduce analog (SCART) ground at 0.5xDVCC.

[zerosignal]

Yes. BTW - do you use RGB or composite?

It's RGB video from the consoles into the switch.

Some great tips, so really thanks. I'll draw up a test plan and look at the datasheets as a start. I'll build another test unit as just having the one is proving difficult to disconnect and get the scope on it.

To be honest I would do tree like structure rather than bus. To build 4:1 switch I would use three 2:1 switches, like MAX9655 which is purpose-built SCART switches IC.

I see what you are onto there, that is an approach I had not considered.

[Renate]

Yes, I meant the SN74CBT6245A? chips as crosspoints.
You might try unsoldering one of these and see how much that channel bleeds into your output.
Especially since all the sources are unsynchronized bleed through will be annoying.

Those are single CMOS switches.
For high quality switching you could use a "T", that is an input series switch, a shunt to ground switch, an output series switch.
The do make that in one package but not sure how many total signals in one package.

I thought about the whole DC value going through.
For game things they are probably all bound by 0 and 5 V and DC coupled, but be sure to check.
They could be AC coupled.
If it were a TV or VCR I would expect some signals below 0.

Normally I'd expect the inputs to be terminated with 75Ω
You didn't show the rest of the circuit so I don't know if the selected circuit is even terminated.

[zerosignal]

The selected input is terminated with 75ohm just before the video amplifier