MItAC update

- 24 hours into parsing 1GB dataset, hit 'OUT OF MEMORY' error on CAML-5:
??? Error using ==> vertcat
Out of memory. Type HELP MEMORY for your options.

- since we do not yet have access to Veeco software calls, we cannot selectively scan so we will start selectively parsing

- documented new bug in Veeco Vision bug blog, bug with Stage File Stitching

- looking into getting an account on CLUMEQ supercomputer http://www.clumeq.mcgill.ca/ and possibilities of speed improvements with our datasets (ie. do we need to parallelize our code?)

MItAC update

- finally managed to verify parser for all test datasets generated from Vision 3.60
- what was thought as unproportional completion times for parser (10MB - 1 minute, 60MB - 1:45min) is actually due to fragmentation of bad data within the Z profile data amongst the good data
- will attempt to parse http://132.206.14.10:16080/mitac/data/78-5-85-3-0-1-5_100ms_raw-float_comp-xyz-pixel_new.asc
without block processing on CAML-5 to see if memory constraints will become an issue. Note that this database is approximately 1GB and CAML-5 has 2GB physical RAM, this may require lockdown of CAML-5 for over 24 hours

MItAC update

- two-point differentiation on old "500ms" images completed (actually only 5ms)
- solved power-on issue (white reset button left of computer should be illuminated to enable power-on)
- installed v3.60, investigated stitching issues (there seems to be size limitations with circular, also we have yet to achieve a "donut")

TO DO:
- investigate stitching and relations to size, time, memory consumption, overlap, etc.

MItAC update

- extracted audio from OLD 500ms databases (this is apparently in reality close to 100ms due to math error)
- extraction without spiral unwrap possible due to negligible curvature of images
- correct playback rate to be determined since area of these images is not known yet

TO DO:
- two point differentiation to model stylus velocity
- solve power issue with microscope
- figure out how to unwrap partial spiral

MItAC update

- documenting all possible turntable imperfections/distortions, posting on Wiki
- summarizing reports on all past work in image/audio conversion (LBNL, Fribourg), posting on Wiki
- July 2006 Wiki update
- received software update, forward Coakley's contact to Ich

TO DO:
- parser is complete, need to finish spiral on partial picture ASAP (absolute position help?), Ich says trace by hand could be acceptable for now
- look into differentiation of path once spiral unwrap complete
- double check Damon's math for 100ms, 500ms measurements on inner groove
- upload code and images to Wiki

MItAC update

- algorithm to locate approximate center of spiral (when entire spiral contained in image) is complete
- due to stitching and memory limitations, currently developing algorithm to locate center of spiral with only incomplete image of spiral
- current approaches will require absolute position data from Vision output header
- we will attempt to extract approx 100ms of sound from innermost grooves due to requirement for less images in comparison with outer grooves

MItAC update

1) created fake spiral data, beginning work on center locating algorithm

2) top-level design appears to match Haber's previous work in 78rpm audio recovery

3) investigation into memory requirements
- please watch http://www.mathworks.com/cmspro/req11290.html?eventid=32518 which describes how to handle large datasets in MATLAB
- 32-bit machine - max 2-3GB, MacOS - max 3-4GB, 64-bit Linux - max 16 GB
- need to consider block processing and image downsampling (textscan, memmapfile)
- data storage - convert from double in Vision to integers
- process - trade speed for memory by devectorizing operations, nesting, using pointers rather than copy write operations between functions, useful tool "monitormatlab"
- plotting - need to downsample

MItAC update

1) new readings and additions to Wiki bibliography ("The Columbia Long-Playing Microgroove Recording System" and "Recording and Reproducing Standards")

2) software top-level design (assuming parser hands over 2D array of Z heights)
a) getbottom.m - normalizes heights and removes data above threshold to leave bottom groove only
b) findcenter.m - locates center of record (whether in array or outside of array) by finding peak correlation with circles of various sizes
c) findgrooves.m - locate grooves along left or bottom axis (start) and indexes their start locations.
d) findspacing.m - determines spacing of grooves from groove locations
e) unwrap.m - unwraps undulations onto the 1D axis

MItAC update

- initial software development plan
1) Parser (Damon)
- goal is to translate ASC dataset into 2D array in MATLAB
2) Setup for audio recovery (Simon)
- goal is to do initial study on technique described below
a) filter top of groove, use only bottom groove and create line of best fit
b) extract single groove from large rectangular stitched image (2D array)
c) fit circular curvature to the groove, extract difference and unwrap groove
d) play sound from displacement from fitted circular curvature

- reported new bugs to Jeff Pageau / Tom Coakley (see http://visionbugs.blogspot.com/)
- software should be shipping out tomorrow from Tucson, wait for a week to speak with Tom about install or call field engineers in Tucson

MItAC update

- preliminary layout and annotated bibliography on Wiki, needs further work on format and will be updated as more papers are read
- further reading into tonearm physics and groove dimensional standards for 78, 33, and 45 rpm
- although summaries and references have been found relating to Columbia intial standard and RIAA standard, will ask for assistance tomorrow to find P. C. Goldmark, R. Snepvangers, and W. S. Backman. "The Columbia Long-Playing Microgroove
Systems," Recording System," Proc. IRE, vol. 37, pp. 923-927
- starting to plan software development for obtaining sound from 78rpm image, reading on vinyl physics ramping down