`Schwartz
`
`[19]
`
`[111
`Patent Number:
`[45] Date of Patent:
`
`4,636,876
`* Jan. 13, 1987
`
`[54] AUDIO DIGITAL RECORDING AND
`PLAYBACK SYSTEM
`
`[75]
`
`Inventor: David M. Schwartz, Denver, Colo.
`
`......................... 381/51
`1/1983 Tahara et a1.
`4,368,988
`4/1984 Henderson et a1.
`381/51
`4,441,201
`
`4,458,110 7/1984 Mozer ......
`331/51
`
`4,519,027
`5/1985 Vogelberg ............................ 381/51
`
`[73] Assignee: CompuSonics Corporation, Denver,
`C010.
`
`Primary Examineerincent P. Canney
`Attorney, Agent, or Firm—Jerry W. Berkstresser
`
`[ * ] Notice:
`
`The portion of the term of this patent
`subsequent to Sep. 18, 2001 has been
`disclaimed.
`
`[21] Appl- No.: 651,111
`[22] Filed:
`Sep. 17, 1984
`
`I63]
`
`Related US. Application Data
`.
`.
`.
`(Izgcgn‘iafit‘glgngafzt 739551" No' 4861561’ AP“ 19’
`’
`’
`’
`’
`
`Int. C1.4 ......... GllB 5/00; GIOL 5/02
`[51]
`
`[52] US. Cl. ............................... 360/32; 381/51
`[58] Field of Search ...................... 381/51, 41; 360/32;
`328/14
`
`[56]
`
`References Cited
`U.S. PATENT DOCUMENTS
`
`[57]
`
`ABSTRACT
`
`.
`.
`.
`A m1crocomputer system for convertmg an analog 51g-
`nal, such as an audio signal representative of sound into
`a digital form for storing in digital form in a highly
`condensed code and for reconstructing the analog sig—
`nal from the coded digital form is disclosed. The system
`includes reductive analytic means .where the original
`digital data stream is converted to a sequential series of
`frequency spectrograms, signal amplitude histograms
`and waveform code tables. Approx1mate1y 100 times
`less storage space than previously required for the stor-
`age of digitized audio signals of high fidelity quality is
`thereby obtained. Additive synthesis logic interprets the
`stored codes and recreates an output digital data stream
`for digital to analog conversion that is nearly identical
`to the original analog signal.
`
`4,164,020
`
`8/1979 Griffith ............................... 364/718
`
`26 Claims, 7 Drawing Figures
`
`_.-
`I
`DATA AOUISITION MODULE (DAM)
`
`'BAND I
`IARRAY
`1
`FERBADT‘.
`5723—51315]
`
`
`1.3 MBYTES / SEC.
`2.6 MBYTES
`(MAX)
`,
`MAX / SEC.
`
`.02 KBYTES/
`(128 CHANNELS)
`SEC.
`
`
`
`
`RAM BUF'FER MODULE
`_. ._ .. ..__|
`,_
`_
`2 KBYTES /
`rAMPLITUDE
`IWAVEFORMS
`I
`|
`1
`
`SEC.\»
`|
`, HISTOGRAMS
`I
`' TABLES
`
`
`I
`1
`l
`_ .1 USEEEXTEEJ
`Lzso xev‘res
`
`
`
`
`
`WAVEFORM ANALYZER
`
`2 KBVTES /$EC.f‘
`8 CODER
`(WAC)
`
`
`130 KBYTES /SEC.
`
`
`(AVERAGEI’N
`
`
` "r—
`DISK RECORD ASSEMBLER
`
`
`
`__1_
`__E___L____‘___E_
`
`NAVEFOR—M— _I
`[71MB
`:
`FREQUENCIES
`IWAVE REFS]
`U514. ______ _I_-_ _ 1.4
`TABLE
`I
`.01 SEC. DISK RECORD
`_1
`[CATALOG
`I
`
`1.... __ ._ _ __________J
`LMAINTENANCE
`
`
`
`
`
`1 I
`
`
`
`MODULE (DRA)
`
`I
`
`WAVE TABLE
`STORAGE
`240.000 .01 SECOND
`RECORDS ON AVERAGE
`
` 130 KBYTES
`/
`
`
`DISKETTE
`MAG, DISK
`STORAGE/
`
`
`
`
`SIGHTSOUND TECHNOLOGIES
`EXHIBIT 2152
`
`CBM2013-00020 (APPLE v. SIGHTSOUND)
`PAGE 000001
`
`
`
`,
`
`PAGE 000002
`
`US. Patent
`
`Jan. 13, 1987
`
`Sheet 1 of7
`
`4,636,876
`
`PLAYER MODULE
`
`,
`
`
`
`I.3 MBYTES / SEC.
`(MAX)
`
`
`DATA AQUlSlTlON MODULE (DAM)
`I.--___,
`I‘___ “I
`I BROAD-I
`H28 BAND
`I BAND
`I
`I ARRAY
`....____._l
`__.__
`
`|I
`
`2.6 MBYTES
`MAX. / SEC.
`
`
`
`
`
`
`' .02 KBYTES/
`
`
`(l28 CHANNELS)
`
`SEC.
`
`
`RAM BUFEER MODULE
`
`__
`_____ _l
`
`
`
`2 KBYTES/
`| AMPLITUDE
`IWAVEFORMS .}
`
`
`
`} TABLES
`SEC.
`,
`I HISTOGRAMS
`I
`
`I
`ZEOJBYIEi
`
`L130. KBYTES
`_._.__ __J
`
`
`
`
`
`
`(WAC) 2 KBYTES /SEC.
`WAVEFORMI ANALYZER
` I30 KBY‘TES /SEC.
`
`(AVERAGE)
`
`
`
`r_____
`
` I
`
`I II J
`
`l L
`
`
`
`,8I CODER
`
`
`DISK RECORD ASSEMBLER /
`MODULE (DRA)
`[—- ——T__'—_' ——— ————— 1—.— -——‘_
`IAMP.
`I
`FREQUENCIES
`lWAVE REFS.)
`IWAVEFORM
`:
`LREF.
`J'_
`_JI
`I-—l
`TABLE
`
`l
`.OI SEC. DISK RECORD
`(CATALOG
`I
`I______ .___________.I
`LII/IAINTENANCE_l
`
`
`
`
`
`
`
`DISK
`READ / WRITE
`
`
`MODULE
`
`
`240,000 .Ol SECOND
`RECORDS ON AVERAGE
`DISKETTE
`
`V
`
`
`
`
`MAG. DISK .
`STORAGE
`
`
`
`WAVE TABLE STORAGE
`
`.ISO KBYTES
`
`FIG.|
`
`
`
`US. Patent
`
`Jan. 13,1987
`
`Sheet20f7
`
`4,636,876
`
`
`
`\\.\“\\%/
`
`
`
`
`
`31vsm / O
`EGnJJ'ldWV
`
`4
`
`PAGE 000003
`
`
`
`US. Patent
`
`Jan. 13, 1987
`
`Sheet 3 of 7
`
`4,636,876 ‘
`
`mm...>m¢
`
`
`
`mo...Qm>mmwwm
`
`
`
`x\\\I/nmzoEaS‘
`
`mDOSmmanowomoomm
`
`Envos us 82
`
`PAGE 000004
`
`
`
`us. Patent
`
`Jan. 13, 1987-
`
`Sheet 4 of 7
`
`4,636,876
`
`
`
`mmEFZmn:muEszo.
`
`
`
`Emfiijwzzxxx
`
`mmEFzmo.
`
`
`
`wn_m<._.EEOmm><>>
`
`mwifizme
`
`
`2m0um><>>
`
`
`
`
`unis.ZmOLm><>>-mfimfi.Emonm><.Em.ImIIIIm—IImmMImm—Immmm
`
`
`a-E-Emir:szz<Io
`IMIIIIEIIEIIMIIIIE
`
`unis.
`
`mmEszn:
`
`amt—2.2mm:
`
`¢.9...—
`
`m0...mostufigdoz<mo<omm
`
`
`
`mqms.Emouw><
`
`
`
`wwomeuummmw4<om
`
`mmfzkzmo.
`
`
`
`u.._m<._.SEOHE?
`
`mm_u_kzuo_
`
`Emma
`
`
`
`
`
`Z>>OIm.rOZwtm>._._m<n_um._.Oz
`
`PAGE 000005
`
`
`
`
`
`
`
`
`
`U.S. Patent
`
`Jan. 13,1987
`
`hS
`
`7:m
`
`
`
`
`
`mMPDn.Zooomuzz«_<zmm._.xm
`
`>JnEDm
`
`mmgom
`
`ZOFUMZZOU
`
`
`
`mqujfiowO43.55
`
`><mm<
`
`.w.5.30528%mamav55xma
`
`
`
`mmmoo<20oz<m
`
`>m02w2
`
`3694x
`
`6
`
`MNm.9“.E05:6.520Big
`
`
`
`..:2:
`
`
`
`$25302;.2525
`
`
`
`
`
`0.5:mjn:2<w¢><mm<
`
`
`
`mN_._._o_o.m
`
`195E024mmm.
`
`
`
`v.03..Smhno
`
`v.92.Saz.
`
`
`
`mjn=2<moz<mo<omm
`
`wN_._._o_ow0.5:
`
`mowmwoomapooUEMEDZ
`
`
`
`mommmoomm»ooOEMEDZ
`
`
`
`mOmmmoommlooDEM—232
`
`
`
`mommwoommyooOEMEDZ
`
`
`
`mgomhzoommmD
`
`PAGE 000006
`
`
`
`
`
`
`
`
`
`
`
`
`US. Patent
`
`Jan.13, 1987
`
`Sheet6of7
`
`4,636,876
`
`
`BAR-CODE
`PRODUC T
`
`TAG
`
`
`
`
`
`IO DUAL-ELEMENT SCANNING
`GUN (BAR-CODE 8‘ RANGE-
`
`FINDER)\
`
`'
`
`
`
`
`
`
`
`DATA VIA
`‘ TELEPHONE LINE
`
`
`
`
`PRINTER
`
`HARDCOPY OF
`FLOOR PLAN 8
`SHELF SPACE
`LAYOUT WITH
`CURRENT INVEN-
`
`_ TORY
`
`FIG. 6
`
`PAGE 000007
`
`
`
`U.S. Patent
`
`Jan. 13,1987
`
`Sheet7of7
`
`4,636,876
`
`
`
`TWME OF SWEEP
`
`
`
` VOLTS
`EDGES OF PRO-
`DUCT fiELD
`
`FREQUENCYINFOR-
`
`MATMN
`
`
`QGNAL PROPORHONAL
`TO DEPTH OF SPACE
`
`
`
`TIME
`
`FIG. 7 ANALOG SIGNAL REPRESENTING SHELF CONTENTS
`
`PAGE 000008
`
`
`
`1
`
`4,636,876
`
`AUDIO DIGITAL RECORDING AND PLAYBACK
`SYSTEM
`
`This application is a continuation-in-part of Ser. No.
`06/486,561, now U.S. Pat. No. 4,472,747 filed 4/ l9/83.
`
`BACKGROUND OF THE INVENTION
`
`Conventional recording of sound and playback is
`performed by electronic systems of the analog type.
`The sound waves from a source being recorded are
`converted to electrical signals on a one to one basis; the
`acoustic sound waves have their analogy in the electri-
`cal current generated by the microphone or pre-
`amplifier circuit such as used in a receiver, turntable or
`magnetic tape source. On playback the electrical signal
`is amplified and used to drive loudspeakers which con-
`vert the electrical signal to sound waves by the mechan-
`ical motion of an electromagnet and speaker cone.
`Similarly, the output of conventional recording and
`playback systems consists of electrical signals in the
`form of signal waveforms either cut into a vinyl me-
`dium or imposed on magnetic particles on tape. On
`playback,
`the signal waveforms are converted into
`sound waves as described above. The accuracy of the
`reproduced sound wave is directly dependent on the
`quality of the metal or plastic disk or of the tape itself.
`Both the production of disk copies and tapes and their
`means of playback tend to degrade the quality of the
`reproduced analog signal. Noise, in the form of contam-
`ination, wear and the inherent background output of the
`medium itself is therefore unavoidably present in the
`recording and playback systems utilizing conventional
`analog to analog recording and playback technology.
`Recent developments in audio-digital sound recording
`and playback systems represent efforts to reduce or
`eliminate this noise problem. Exemplary of such devel-
`opments are the kinds of systems and equipment dis-
`closed in the following patents: Meyers et a1, U.S. Pat.
`No. 3,786,201 issued Jan. 15, 1974; Borne et a1, U.S. Pat.
`No. 4,075,665, issued Feb. 21, 1978; Yamamoto, U.S.
`Pat. No. 4,141,039, issued Feb. 20, 1979; Stockham, Jr.
`et a1, U.S. Pat. No. 4,328,580 issued May 4, 1982;’Tsu-
`chiya et a1, U.S. Pat. No. 4,348,699 issued Sept. 7, 1982;
`and Baldwin, U.S. Pat. No. 4,352,129 issued Sept. 28,
`1982, the disclosures of which are specifically incorpo-
`rated herein by reference. These systems are character-
`ized generally as taking advantage of the high speed
`operation of the digital electronic computers. The sig-
`nal waveform, representative of sound in such digital
`sound recording and playback systems,
`is frequently
`sampled to produce a serial stream of data that is trans-
`lated into a binary code that assigns a numerical value
`for each sample. This can be visualized as slicing up a
`continuous curve into a large number of very short
`step-like segments. The process is reversed on playback
`as each numerical value of each segment is converted
`into an output voltage. When this process is done rap-
`idly enough, the fact that the signal wave form repre-
`sentative of a sound wave has been “chopped up” and
`re-assembled cannot be detected by the human ear.
`When sound is recorded in digitized binary code in this
`manner, the sound, such as music, is only a series of
`numbers represented by magnetic tracks on a recording
`medium which, when read by the appropriate elec-
`tronic means, are either “on” or “off" with no interme-
`diate values. Such binary signals are virtually immune
`to distortion, error, and degradation with time. All
`
`5
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`45
`
`50
`
`55
`
`6O
`
`65
`
`2
`sources of noise normally associated with analog de—
`vices are eliminated that is, there is no tape hiss, no
`tracking errors, no surface effects. Signal to noise ratios
`are limited, only by the digital to analog conversion
`circuit itself and the power amplifiers, rather than the
`sensitivity of the mechanical or magnetic analog to
`analog conversion circuitry.
`These systems do, however, have several drawbacks.
`A representative system currently in use for recording
`master tapes in the record industry has excellent audio
`qualities as a result of a high speed sampling rate of 50
`KHz and good digital binary code resolution in the
`form of a 16 bit word for each sample. The problem
`with this system is that every sample must be preserved
`in mass storage for playback. The storage system thus
`must hold on the order of 4,320,000,000 bits of informa-
`tion for a 45 minute record. Storage systems of this
`capacity are large, expensive, and generally not suitable
`for a consumer product.
`Attempts to resolve the storage capacity problem
`have taken the approach of reducing the resolution of
`each sample (fewer bits per “word”) while at the same
`time reducing the sampling rate (to 12 khz). Such reduc-
`tions have reduced the data storage requirement by as
`much as a factor of 4. The resulting fidelity of the out-
`put, however, is often below that acceptable for high
`fidelity sound recordings of music.
`Another approach much favored by telephone com-
`panies, employs the foregoing reduction of bits de-
`scribed above and in addition adds the restriction of
`input signal band width to that most used by talking
`voices (3 to 8 KHz). A total data reduction factor of
`about 12 is possible in this manner, again accompanied
`with a reduction in sound quality.
`Recent attempts at a solution to the storage problem
`and the fidelity reduction problem utilizes ultra high
`density digital storage by laser recording technology.
`This has been partially successful in that adequate .play-
`ing times have been achieved with the improved stor-
`age capacity. However, the manufacturing technology
`and equipment presently necessary to create a “laser-
`burned hole”, “pit”, or “black spot” in the storage me-
`dium restricts “laser disks” or “laser fiches” to the
`“playback only” mode with no potential for in-home
`recording or erasing and editing.
`It is therefore an objective of the present invention to
`provide a system for high fidelity sound recording and
`playback that does not have the foregoing drawbacks
`and associated problems.
`SUMMARY OF THE INVENTION
`
`The present invention is yet another approach to a
`solution to the storage and reproduction problems asso-
`ciated with digital audio recording and play back sys-
`tems described herein. Good fidelity can be achieved
`with limited computer storage capacity by the provi-
`sion of unique electronic signal processing means
`which: (1) converts analog data to digital data stream
`samples; (2) selects portions of the samples to produce
`at least three data streams indicative of amplitude, fre-
`quency and waveform characteristics; (3) stores data
`samples indicative of waveform having a predetermined
`time duration, comparing each such sample of wave-
`form data against predetermined waveform parameters
`to select and preserve only predetermined portions, said
`waveform data samples matching the preserved por-
`tions with pre-existing waveform and real time data and
`generating a resultant waveform data code from such
`
`PAGE 000009
`
`
`
`3
`comparison, and then comparing the selected data from
`the data streams which are indicative of frequency and
`amplitude with the waveform data code to produce
`another data code proportional to the frequency and
`amplitude of the original analog signal, sequentially
`recording the data stream indicative of amplitude, the
`data code indicative of frequency and amplitude, and
`the data code indicative of waveform, onto a recording
`media, for subsequent playback by the processing of the
`sequentially recorded data.
`BRIEF DESCRIPTION OF THE DRAWINGS
`
`FIG. 1 is a schematic diagram of the digital recording
`and playback systems of the present inventiOn.
`FIG. 2 is a pictorial representation of the analytical
`model of the function of the Data Acquisition Module
`of FIG. 1.
`FIG. 3 contains a diagramatic representation of the
`recorded waveform data.
`FIG. 4 is a pictorial representation of a single unit of
`binary code as stored on disk, from which reproduction
`will be obtained according to the system of the present
`invention.
`FIG. 5 is a diagramatic representation of the layout of
`the electronic components used in the present inven-
`tion.
`
`FIG. 6 is a pictorial representation of a warehouse
`inventory system.
`' "FIG. 7 represents an analog signal output of the appa-
`ratus of FIG. 6.
`
`DETAILED DESCRIPTION OF THE
`INVENTION
`
`The present invention provides a system for convert-
`ing input analog signals such as audio signals into digital
`signals and subsequently coded into structured data sets
`for recording in condensed digital form; and, for recon-
`structing a digital data set similar to the original digital
`fl signal input prior to reconversion to the analog form of
`, signal.
`In its broadest sense, therefore, the recording of the
`audio signals into a digital form for subsequent playback
`is accomplished by the provision of ‘a microcomputer
`recording system which comprises electronic compo-
`nents for converting an analog audio signal into at least
`three digital data streams, wherein the first of the digital
`data streams is a relatively broad band reference signal
`representative of the amplitude of a pre-selected range
`of audio frequencies, and the second of said data streams
`is produced by filtering the analog audio signal to pro-
`duce at least one data stream channel indicative of a
`sampled band width of frequencies narrower than the
`band width represented by such first data stream, and a
`third reference data stream representative of the sam-
`pling frequency of the audio signal; sampling means for
`producing a sequential stream of data samples from
`each of the digital data streams, selection means for
`selecting a pre-determined portion of the digital data
`sample produced by the sampling means in each of the
`data streams; means for separately storing each of said
`selected digital data samples produced by the sampling
`means; means for comparing the reference signal data
`stream containing amplitude data with said second data
`stream containing frequency data to produce frequency
`spectrogram data representative of the frequency and
`amplitude, of the original audio signal; means for trans-
`forming data samples of the third data stream channel
`selected from the narrower band width into data repre-
`
`'
`
`5
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`4O
`
`45
`
`50
`
`55
`
`6S
`
`4,636,876
`
`4
`sentative of a time versus amplitude histogram for each
`band width; means for comparing said histogram data
`with selected waveform parameters and producing and
`storing addressable data representative of the waveform
`of the original audio input and means for sequentially
`assembling and storing the frequency spectrogram data
`and the amplitude reference data of the first data stream
`and the addressable waveform data for subsequent play-
`back use.
`
`In the preferred embodiment shown in FIG. 1, the
`input signal
`is conditioned and amplified in the first
`stage of the Data Acquisition Module (DAM). The
`DAM is a multi-channel programmable microprocessor
`based device that utilizes standard integrated circuits to
`perform three functions:
`1. To sample at the rate of 42 Khz, hold, digitize, and
`output the broadband (20 hz to 20 Khz) audio signal
`level (dc voltage) of amplitude every 0.01 second. Thus,
`100 times every second a digital “word” composed of
`from 4 to 14 bits is created for assembly as part of a disk
`record file.
`
`2. To sample, hold, digitize and output an audio fre-
`quency spectrogram every 0.01 second from a 128 seg-
`ment array of logical bandpass filters which sample 128
`channels and are arranged logarithmically over the
`overall band width used. The data set produced by this
`function may range from null (no signals on any chan-
`nel)
`to (n)
`[(7 bit
`identifier+(7 bit scaler)+(2 bit
`pointer)] where (n) is the number of channels with sig-
`nal content.
`3. To act as a digital storage oscilloscope loader,
`assembling strings of digitized amplitude versus time
`data (histograms) corresponding to the array of band-
`pass filters selected in paragraph 2, above. This assem-
`bled data set is produced every 0.01 second and is the
`largest single data structure and contains time continu-
`ous listing for every active bandpass filter. The number
`of “words” in each string is a function of the filter cen-
`ter frequency and requires as many as 4,000 samples for
`a 20 Khz channel, or as few as five samples for a 20 hz
`channel. This data set is not sent to the file assembler as
`in paragraphs 1 and 2, above, but is loaded into a Ran-
`dom Access Memory (RAM) buffer where it is accessi-
`ble by the Waveform Analyzer and Coder module.
`The function of the Waveform Analyzer and Coder
`module (WAC FIG. 1) is to be a digital numeric proces-
`sor array that is programmed to extract characteristic
`wavcforms from the data set stored in the RAM by the
`DAM described above. The waveform data are re-
`duced to tabular form in which one period of each
`waveform Codified is assigned to one wave table which
`preferably is a digitized x-y coordinate system consist-
`ing of 1,024 bytes along the x axis and an 8 bit “word”
`in each byte location to scale the y axis in 256 incre-
`ments; 127 above zero and 127 below. A set of waveta-
`bles is therefore generated for all active bandpass filter
`channels every 0.10 second. A range of 0 to 128 P.M.S.
`tables may be generated per cycle (0.01 second).
`The WAC utilizes either one of several P.M.S. reduc-
`
`tive analytic methods to find waveforms. The first being
`the Fast Fourier Transform (FFT) and the second the
`Fast Delta Hadamard Transform (FDHT). The two
`methods may be briefly described as follows:
`The FFT is based on the principal that almost any
`periodic function f(x) of period 2 of vibrations can be
`represented by a trigonometric series of 'sines and co-
`sines. The full expression in general terms is:
`
`PAGE 000010
`
`
`
`4,636,876
`
`6
`The Waveform Catalog Maintenance Subroutine is
`programmed to evaluate incoming updates of wave-
`form tables against
`the waveform tables previously
`stored, and among themselves. Since there are only 128
`channels available for storage of the amplitude histo-
`gram output of the DAM, the comparison of the wave-
`form output of the WAC with the stored waveform
`data of the DRA determines redundancy and duplicates
`are discarded. The remaining incoming tables are possi-
`bly unique or simply improved versions of forms al-
`ready stored. Waveforms that contain new features on
`existing tables are saved in place of their previous table
`resident. Unique forms are assigned new tables. When
`an overload occurs due to a “full house” and a unique
`waveform arrives it is placed in a local buffer for one
`cycle to determine its repetitiveness. If indeed it does
`occur in the next 0.10 second cycle, space in the Wave-
`form Catalog is made for it by discarding the waveform
`most similar to another in the Catalog. The algorithms
`used for these evaluations are based on standard statisti-
`cal methods for measuring the fit of two curves to one
`another.
`
`w
`" co
`
`00
`— do
`
`.
`j(v)e'”(x‘ V)dv
`
`dw
`
`The algorithm for executing this equation efficiently
`was first published by Rabiner & Gold, 1974 and Op-
`penheim and Schafer, 1975.
`The FDHT is utilized for the analysis of spectral
`composition of a data set where the spectrum ‘11 is in the
`form:
`
`1:
`1110') = '5; i8[1-‘— (FI'+ 1 + Fz)/z]
`
`where Fi is the frequency and ‘11 i is signal intensity. In
`the present application of this method the digital output
`of the logical filters from hereinbefore numbered para-
`graph 2, is summed at each filter and added to the next
`output until all frequencies have been sampled. At the
`last step the total output is:
`
`Then an estimation of the spectrum (W') can be found by
`matrix multiplication:
`
`1
`1
`,
`1p =—n—SB.-Q=TSB.S.¢=¢
`
`The algorithm for implementing the FDHT was pub-
`lished in 1983 by E. E. Fenimore at Los Alamos Na-
`tional Laboratory.
`B—splines computational algorithms may also be em-
`ployed to extract characteristic waveforms.
`Ten times every second the latest produced set of
`waveform tables are sent to the Disk Record Assembler
`(DRA FIG. 1).
`'The Disk Record Assembler (DRA) is a program
`module that receives as input the waveform table refer-
`ences (addresses) from the WAC every 0.10 seconds
`and paragraph 2 (above) frequency spectrogram data
`sets every 0.01 seconds directly from the Data Acquisi-
`tion Module (DAM) as well as the digital word repre-
`senting the total broadband signal strength. The wave-
`form tables are kept in a local memory buffer in the
`DRA so that they may be revised or discarded every
`0.10 second cycle by a subroutine which for conve-
`nience will be called Waveform Catalog Maintenance.
`Disk records (FIG. 4) for storage are variable in length
`but always follow this format: the first 14 bits are the
`field length statement, the next 7 bits are the frequency
`filter or channel identifier followed by a 2 bit pointer
`(flag) and its 7 bit sealer, 7 bit waveform table identifier,
`7 bit simultaneous waveform table identifier (repeat if
`necessary), 2 bit flag (signals next filter identifier), and
`so forth to the last 14 bit word which is the broadband
`signal level. The data stream format is shown graphi-
`cally in FIG. 4.
`Once a record is prepared for storage it is held in a
`local memory buffer in the DRA for one cycle so it can
`be compared to the next sequential record. This allows
`the DRA to utilize “tokens”; specific reserved symbols
`to identify “repeats”, “same excepts” and “nulls” in the
`current record being assembled to save storage space.
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`40
`
`45
`
`50
`
`55
`
`60
`
`65
`
`In the preferred embodiment of this invention the
`storage medium is a 5.25” magnetic disk commonly in
`use for digital magnetic storage and retrieval. These
`disks have a storage capacity of about 1 megabyte (1
`million bytes or 8 million bits) and are anticipated to
`reach 10 megabytes in the near future. For purposes of
`illustration, a 5 megabyte disk will be assumed.
`Assembled disk records from the DRA are the input
`for the Disk Read/Write module. In the “write” mode,
`records in the form of the data stream format previously
`described, will be written to disk storage as long as
`there is space available. Considering an average record
`to be 20 bytes of data the disk will contain about 240,000
`records, each representing 0.01 seconds of real time. In
`addition the entire Waveform Catalog is written to disk
`after all space on the disk has been filled except for the
`130 Kilobytes required for the Waveform Catalog itself.
`In the retrieve mode, or playback, the Disk Read/—
`Write Module first reads the Waveform Catalog from
`the disk into RAM. The waveform tables are then ac-
`cessed by the Player module when called within each
`disk record. Each 0.01 second disk record is read from
`the disk serially to preserve its relationship to the real
`time of the original audio source material.
`The Player module utilized in the present invention
`will preferably contain digital oscillators to produce the
`output signal and “smoothing” filters to eliminate the
`“steps” inherent in digital representations of continuous
`analog functions. Additive synthesis is the principal
`upon which the Player module’s logic is based. Briefly
`summarized, additive synthesis theory states that com-
`plex musical signals can be constructed by summing all
`of the voltage components of the signal of each fre-
`quency subset at each instant in time. Thus, if the data
`reduction process preserves all of the original informa-
`tion about voltage versus time in such a way that it can
`be recombined precisely and in phase in time the output
`signal will equal the original input signal in each speci-
`fied frequency or “pitch”. In the preferred embodiment
`of the invention these conditions of additive synthesis
`are preserved at a level of perceptual resolution so that
`what the human ear will hear is indistinguishable from
`the original for mest source material. The Player mod-
`ule then directs the oscillators to output at the frequen-
`cies specified by the disk records utilizing the waveform
`reference data to set the timbre of each oscillator and
`
`PAGE 00001 1
`
`
`
`4,636,876
`
`7
`the broadband amplitude reference data sets the voltage
`levels. Synchronized timing is built into the system by
`definition of the 0.01 second cycle time upon which the
`system is based.
`A most preferred embodiment of the system will
`employ Very Large Scale Integrated Circuit (VLSIs)
`technology to reduce logical groupings of circuit to
`single semiconductor chips, as opposed to the schematic
`representation shown in FIG. 5 which utilizes many
`“off the shelf” Integrated Circuit components.
`Referring now to FIG. 2 the analytic model is graphi-
`cally depicted. The model has three reference axis di-
`mensions of measurement; time, amplitude (dc voltage),
`and frequency. The time axis is divided into 0.01 second
`increments. It is important to the understanding of the
`system of the present invention to realize that the 0.01
`second interval corresponds to the rate at which incre-
`mental acoustic “snapshots” of the audio signal are
`recorded. This increment was chosen because it is short
`enough that the human ear physiologically hears a se-
`quence of 0.01 second changes in total signal as a con-
`tinuous integrated whole. The stream of acoustic “snap-
`shots” is directly analogous to the stream of “frames” in
`a motion picture film.
`The acoustic “snapshots” themselves contain, in bi-
`nary form, the total broadband (20 to 20,000 hz) ampli-
`tude, a frequency spectrogram and waveform table
`references obtained from the DAM (FIG. 1). The illus-
`v....tration in FIG. 2 amplitude histograms, such as (ahé)
`shows the waveforms contained in the so-called “ampli-
`tude histograms” which are the raw data sets used to
`,» write the waveform tables. This will be discussed in
`greater detail hereinafter. The total broadband ampli-
`tude record is the reading, every 0.01 seconds, of a
`continuous digital stream of 14 bit words “written” by
`the broadband sample, hold and digitizing circuit at the
`rate of 42,000 “words” per second. Viewed another
`. way this is like saying that only one word is saved for
`every 420 created. This series of amplitude readings is
`1.; utilized from the RAM Buffer Module in the “playing”
`of the digital oscillators at the output end of the system.
`Every amplitude reading in every frequency channel is
`scaled to this reference level. Referring again to FIG. 2;
`(BER) “broadband reference record” is a 2 dimensional
`data array in which the first term is the time value
`within the 0.10 second time frame incremented every
`0.01 seconds (i.e. 0, 0.01, 0.02, seconds). The second
`term is the binary representation of the dc voltage level
`or amplitude at each time increment. The voltage level
`is recorded to the accuracy of a 14 bit word. This allows
`16,384 discreet values for representation of the dc volt-
`age range which may typically be from 0.05 volts to 5
`volts i.e. 100 db. The absolute accuracy is thus 4.95
`divided by 16,384 or $00003 vdc.
`It would be desirable to have this level of accuracy
`for the vdc measurement recorded in each bandpass
`filter channel. However, to achieve economy of storage
`space it is desirable to use as few bits as possible to
`represent the amplitude of the signal in each channel.
`To accomplish these contradictory goals the method of
`relative representation is adopted. Each frequency
`channel amplitude record is a bit word called a scaler
`value, that allows 128 values, which records each chan-
`nel’s signal as a proportion of the broadband value.
`Thus a channel with a vdc that is 0.250 vdc when the
`broadband value is 3.250 has a proportional value of
`0.07692 with respect to the broadband signal. On a 7 bit
`scale this is a “3” out of 128. The second benefit of this
`
`.
`
`8
`
`10
`
`15
`
`20
`
`25
`
`30
`
`35
`
`45
`
`50
`
`55
`
`60
`
`65
`
`8
`approach is the increased speed of computation af—
`forded by the comparative nature of modular arithmetic
`logic as opposed to the more time consuming logic for
`accumulating and encoding a 14 bit accurate “word” at
`each channel, thus utilizing a 7 bit word instead of a 14
`bit word is a 50% savings in storage space.
`Referring now to FIG. 3, the frequency spectrogram
`Fs 10 is similar to the broadband amplitude record ex-
`cept that the amplitude of the voltage in each of 128
`discreet narrow bandwidths is “saved” every 0.01 sec-
`onds. The 128 channels are sample, hold and digitizer
`circuits that are limited to the bandwidths they can
`“hear” by preselected digital bandpass filters. The dis-
`tribution of the channels across the 20 to 20,000 hz
`audio range may be controlled by the user with equal-
`izer type slide switches or may be automatically signal
`seeking. That is, the logic of the 128 channel array
`responds to the amplitude of the voltage in each of 128
`discreet narrow band widths and “self~centers” around
`“live” bandwidths. This principal is the same as used in
`signal seeking radio receivers.
`_
`As representatively shown in FIG. 2, there can be
`overlaps between channels such as shown by the shaded
`triangular regions on the frequency spectrogram axis. A
`signal in the overlap region indicates to the system logic
`that the channel array is not “in tune” with the incom-
`ing digitized signal and can serve to set a flag value for
`correction that can be used by the automatic ranging
`circuit to “step over” to the next acoustic “snapshot” to
`get a centered channel reading.
`The amplitude histograms , for example ah6 in FIG. 2,
`are created whenever a channel is “live”. These histo-
`grams are point by point amplitude versus time binary
`plots that are generated on a real time continuous basis.
`They are not 0.01 second “snapshots”. The actual
`length in time required for plotting a histogram will
`vary with the audio frequency of the channel. It is gen-
`erally conceded that the higher the frequency, the more
`data points will be required to “feed” the Waveform
`Analyzer and Coder. Of course, the upper limit in time
`for this process is 0.10 seconds or the synchronization of
`the entire system would be affected. The purpose of the
`amplitude histograms is to provide the “raw data” for
`the FFT or FDHT routines that operate the WAC. In
`order for the FFT to characterize a series of X-Y coor-
`
`dinates as a periodic curve function at least 2 complete
`cycles of the periodic function must be collected. In
`many cases, due to electronic recording logic circuit
`delays often referred to as “settling time” disturbances,
`more than 2 cycles worth of data must be collected for
`analysis.
`Referring now to FIG. 3, the Wave Table Catalog
`information is graphically represented in its preferred
`form for the system. As soon as the Waveform Analyzer
`and Coder (FIG. 1), has “found” a waveform in an
`amplitude histogram (FIG. 2 aha) the waveform data
`for one period of the waveform is plotted on an X-Y
`coordinate system as shown graphically in wave table
`wt1, of FIG. 3. The amplitude of the wave is plotted in
`the y dimension with 1,020 8 bit binary words that allow
`a precision of 127 steps above and below the x axis. The
`x axis itself is an arbitrary division of the waveform’s
`period into 1,020 increments. The wave table has four
`bytes reserved for information about the tables status
`within the catalog of 128 tables. This is necessary since
`references to wave tables po