# Notes First photogrammetry to get a virtual model of the space Met conditions too for use with https://sengpielaudio.com/calculator-airpressure.htm to get speed of sound. Sync with sound at start, same at end to prevent clock drift. 5 different positions to reconstruct. Project freq: 44100 Hz (that gives you a specific precision) Need to do it manually cause detection is at the sample level and complicated to calibrate accurately since shorter durations at play. I chose this mic disposition to show that error is concentrated where the spheres are parallel to each other (the claps that are close to the sort of plane created by the mics are really precise whereas as soon as you move away from the plane you lose a lot of precision) # Samples (Left Channel Only (Top in Audacity)) (plane formation) Start: 2106421 (each input aligned at start) ## Computer mic (atomic clock reference within 0.62ms root dispersion) Clap 1: 5226608 Clap 2: 5408665 Clap 3: 5599824 Clap 4: 5759440 Clap 5: 5879540 End: 7134747 ## Xiaomi mic Clap 1: 5226652 Clap 2: 5408683 Clap 3: 5599858 Clap 4: 5759346 Clap 5: 5879681 End: 7134723 ## Google Pixel mic Clap 1: 5226677 Clap 2: 5408768 Clap 3: 5599835 Clap 4: 5759503 Clap 5: 5879506 End: 7134793 ## Camera mic Clap 1: 5226654 Clap 2: 5408705 Clap 3: 5599821 Clap 4: 5759352 Clap 5: 5879593 End: 7134678 # Results (plane formation) Clap 1 (error): 0.3080m Clap 2 (error): 0.5229m Clap 3 (error): 0.0972m Clap 4 (error): 0.6249m Clap 5 (error): 0.3638m In the detection plane we get about 10cm of error and outside about 50cm of error. This is consistant with the calculated point cloud with standard deviation of 2 samples per detector error (calculated with Monte-Carlo). (Adjusted the standard deviation of 2 samples per detector error with clap 3 to have about 10cm of error at its real location and that indeed gives you about 50cm at the other locations) # Samples (Left Channel Only (Top in Audacity)) (symmetric formation) Tried again this time with a symmetric formation Start: 1253385 (each input aligned at start) ## Computer mic (atomic clock reference within 0.62ms root dispersion) Clap 1: 2718755 Clap 2: 2886568 Clap 3: 3093112 End: 4036375 ## Xiaomi mic Clap 1: 2718813 Clap 2: 2886539 Clap 3: 3093112 End: 4036348 ## Google Pixel mic Clap 1: 2718843 Clap 2: 2886703 Clap 3: 3093149 End: 4036401 ## Camera mic Clap 1: 2718799 Clap 2: 2886606 Clap 3: 3093109 End: 4036372 # Results (plane formation) Clap 1 (error): 0.2201m Clap 2 (error): 0.2039m Clap 3 (error): 0.1499m About 19cm of error. Less variance across the entire space. Error is consistant with simulations too.