It's... definitely a filter. I can say that much.

Not sure what's up with the 0 dB regions at start and end of the sweep, it's clearly a bug (The waveform behavior doesn't match, it's definitely not actually 0 dB)

I'm pushing limits of what I can do as far as dynamic range easily. The Siglent is putting +16 dBm into the doubler (which has an absolute max of +20 so it's safe but not leaving much margin). The doubler is only specced for 10-20 GHz input so 5-10 GHz input and trying to sweep down to 4 GHz on the input means I'm operating out of spec but it seems to mostly work.

The doubler has ~12 dB of conversion loss so I only have about +4 dBm coming out of it, which isn't a ton of stimulus energy when the DUT is this lossy.

I'm not sure why it's coming out as 30 dB IL. My simulations said it would be more like 15, plus copper roughness and ENIG plating losses which I did not expect to add another 15 dB.

I expect the filter rolloff is a lot sharper than what's shown here but the -45 dB loss range is putting me basically at the noise floor of the scope.

I could try to get more dynamic range by adding an amplifier after the doubler, but i'm a bit sketched out about putting big RF PAs on something going into a $10K-to-repair-if-damaged 16 GHz frontend that is rated for +/- 2V p-p.

Probably better to just send the thing to somebody who has a real VNA that can work at these frequencies.

@nprofile1q... what are they doing with that? a few tens of kilograys are what's typically used in industrial sterilization lol

@nprofile1q... i looked. According to the sales rep i talked to they do not have any chemical resistant label/ink offerings otherwise they looked attractive

@nprofile1q... yeah so far nobody has everything i want but i can tolerate windows software in a VM in a pinch.

Brady seems the only option for chemical resistant full color that I've been able to find

@nprofile1q... doesn't appear to be full CMYK color

@nprofile1q... I'm currently planning to spend $150 on an eBay J2000 to keep me staggering on a bit longer and burn up the remainder of my (expensive) consumables, then buy a J7300 when i run out

Since the vast majority of libscopehal shaders are memory bound, if I can manage to fuse several shaders into a single one and keep the intermediate results in registers, there should be a massive improvement in performance.

Fusing just two consecutive 1:1 shaders will halve the memory bandwidth requirement of the pairing.

Tl;dr identifying dependency chains where two shaders are called back to back without the intermediate result being used anywhere else, and the sample processing is either strictly 1:n or only works in a small window around the input, like a FIR filter below a few dozen samples in size or something.

Then fuse the GLSL for each kernel into a single combined shader, JIT to SPIR-V, and make a single compute shader pipeline from it.

@nprofile1q... i dont even know if this is a shitpost or if IoT disposable culture has jumped the shark this bad

@nprofile1q... My general opinion on this sort of thing is, it's a tool for when you'd rather take the charge for destruction of evidence instead of the charge for whatever they'd find otherwise is.