Compact Cassette and DAT as Data Media (1963-2009)

Short answer: audio tape became a data medium twice. The Philips Compact Cassette (shown in August 1963) was turned into a computer store by the Kansas City standard of November 1975, which recorded bits as 2400 Hz and 1200 Hz tones at 300 baud and was soon replaced in practice by each home-computer maker's own faster format; and the same 3.81 mm tape width came back in 1987 as consumer Digital Audio Tape (DAT), whose rotating-head, helical-scan drive was adapted by HP and Sony into the Digital Data Storage (DDS) backup format of 1989, which grew from 1.3 GB to 160 GB native over twenty years.

1963: the cassette. The Philips Museum article says the recorder development group at Philips in Hasselt, Belgium was commissioned in 1961 to develop a small battery-powered recorder and cassette for home use, and that the Pocket Recorder EL 3300 and cassette EL 1903 were first shown at the Internationale Funkausstellung in Berlin in 1963. It says Philips gave other manufacturers a free licence provided they followed the technical rules, so that the Compact Cassette became the world standard; the museum's own timeline (a 2010 Wayback capture) says the EL 3300 ran at 4.75 cm per second and that no licence fee was charged. A 2013 interview with Lou Ottens quoted in The Register (search excerpt only) says that CBS's Peter Goldmark had proposed a narrower tape of 0.15 inch (3.81 mm) at 1 7/8 inch per second (4.75 cm/s), and that Ottens said Philips kept to existing standard tape speeds. The museum article says the recorder was first intended for recording spoken word and was widened to music within months, so data use was never part of the design.

1975: the Kansas City standard. BYTE magazine held a symposium on 7 and 8 November 1975 in Kansas City, Missouri, to agree how to interchange data on 'inexpensive consumer quality audio cassette drives' (BYTE, February 1976, pp. 72-73). The provisional standard, written up by Lee Felsenstein and Harold Mauch, defines a mark (logical one) as eight cycles of 2400 Hz and a space (logical zero) as four cycles of 1200 Hz; a character is a space start bit, eight data bits least significant first and two or more mark stop bits; data blocks are preceded by at least five seconds of marks and begin no sooner than 30 seconds from the start of clear leader; and the nominal tape speed is 1.875 inch per second (4.75 cm/s). BYTE says the technique is 'a redundant form of Manchester or bifrequency code', assumed a reliable bandwidth of 3000 Hz because of expected head misalignment between recorders, and that bit clocking can be recovered whatever the tape speed, so a UART could decode it. The 18 attendees listed include Lee Felsenstein and Bob Marsh of Processor Technology, Ed Roberts, Tom Durston and Bill Gates of MITS, and Gary Kay of Southwest Technical Products. At 300 baud with an 11-bit frame the standard gives 300 / 11 = 27.3 bytes per second (calculation; the same figure is in a Retrocomputing Stack Exchange question and in the Wikipedia article), or about 24.5 kB on a 15-minute side before any block overhead (calculation).

Names and variants. A Retrocomputing Stack Exchange answer says Processor Technology's Computer User Tape System (CUTS) is 'more or less a synonym' for the Kansas City standard but also defines a 1200 baud mode that the standard did not adopt, and that Acorn's '1200 baud Kansas City' format used the same tone frequencies with different framing. The Wikipedia Talk page (search excerpt) disputes the claim in the main article that the standard was published in BYTE's first issue: it says the September 1975 issue printed Don Lancaster's proposal, and the standard itself was printed in February 1976, which matches the BYTE page read for this card. This card does not resolve the CUTS naming question beyond what those two sources say.

Late 1970s and 1980s: every machine its own format. The Retrocomputing Stack Exchange question and its answers give a table of nominal rates, none from a manufacturer document: TRS-80 Model I 250 baud (Level 1) and 500 baud (Level 2), ZX81 300, Atari 400/800 600, Amstrad CPC 1000 or 2000, Apple II and IBM PC about 1500, ZX Spectrum about 1500, Acorn 300 or 1200. The trs-80.org and GitHub excerpts agree on 250 and 500 baud for the Model I and add that the Model III's 1500 baud mode is frequency-shift keying with a 1320 Hz cycle for a zero and a 2680 Hz cycle for a one, where the Model I uses pulses (the GitHub author measured 1508 baud on one program). For the BBC Micro, one Stack Exchange answer works out 1200 baud FSK at 1200 and 2400 Hz with 256-byte blocks, about 66.2 bytes per second sustained and 'a little under 60 KB' on a 15-minute side (single answer, calculation by the answerer).

The Commodore Datassette. Wikipedia says the Commodore 1530 (C2N) Datasette was Commodore's dedicated cassette unit for the PET, VIC-20 and C64, that the earliest 1977 PET models had a modified Sanyo cassette drive built in, and that the unit sends digital signals to the tape rather than audio. The encoding is pulse-length, not tone: three pulse lengths of 176, 256 and 336 microseconds on NTSC machines (183, 266 and 349 microseconds on PAL), with a bit as a short-medium or medium-short pair, a byte marker as long-medium, and an odd-parity bit (C64-Wiki, citing Nick Hampshire's The PET Revealed, 1980; Wikipedia, citing Compute!, 1985). Both give 8.96 ms per byte (C64-Wiki: 'each byte has a recorded duration of 8.96 ms'), so the raw rate is about 111.6 bytes per second (calculation; also stated in a Stack Exchange answer). Wikipedia's own encoding section also says 4480 microseconds per byte and 223.2 bytes per second, which conflicts with its own 8960 microsecond figure and with C64-Wiki; this card uses 8.96 ms. Each 192-byte block is stored twice for error recovery (C64-Wiki), so the sustained rate is roughly half the raw rate: Wikipedia says about 50 bytes per second, a Stack Exchange answer about 55. The capacity of a side is also disputed: Wikipedia says 'about 100 kByte per 30 minute side', the Stack Exchange answer calculates about 48 KB for a C30 side with the standard format.

Fast loaders. Wikipedia and a Stack Exchange answer both say that third-party 'turbo' loaders for the C64 sped loading by dropping one of the two pulses per bit and dropping the per-byte markers and parity. Wikipedia adds that turbo tape approximately doubled bit density and says a side then held 'roughly 1000 kByte', a figure that is not consistent with the answer's calculation of a typical 2778 bit/s fast loader giving about 347 bytes per second and about 300 KB per C30 side; the two are shown as conflicting and neither was checked against a primary source. Wikipedia says the Datasette was more popular in Europe than in the United States, citing Compute!'s Gazette (1983), Computer Gaming World (1986) and a MicroProse statement (1987), and that US buyers mostly bought disk drives; that comparison is single-source (Wikipedia citing magazines not opened here).

1983-1987: from audio cassette to digital audio tape. Audio magazine (July 1987) says a conference on digital audio tape recording was organised in mid-1983 and eventually involved about 84 companies; tentative R-DAT (rotary head) and S-DAT (stationary head) specifications were compiled in June 1985, and the final R-DAT standard was issued in June 1986. The same article says DAT recorders had been on sale in Japan since the third week of March 1987 and were not yet sold in the United States. Stereophile's specification sidebar for the Sony DTC-1000ES gives 3.81 mm tape, a 30 mm drum at 2000 rpm, tape speed 8.15 mm/s, head-to-tape writing speed 3.133 m/s, 90 degree wrap, azimuth angle of 20 degrees, two-hour recording time, 16-bit linear quantisation, a data transmission rate of 2.46 Mbit/s and dual Reed-Solomon error correction; a Hifi-Wiki excerpt calls it the first DAT recorder available on the free market. Audio magazine adds that tape is wrapped around a quarter of the drum, successive tracks have different azimuth so no guard bands are needed, and that a C90 analog cassette holds about 420 feet of tape for 45 minutes per direction while the DAT needs less than 100 feet for two hours. The 3.133 m/s writing speed is about 384 times the 8.15 mm/s tape speed (calculation), and about 66 times the 4.75 cm/s cassette speed (calculation). The DAT track also carries a subcode area whose copy-prohibit flag, per Audio magazine, stopped digital-to-digital copying of flagged sources.

1987-1989: DAT becomes DDS. StorageNewsletter's abstract of a May 1989 trade newsletter says HP and Sony announced a joint development in August 1987, that HP's Bristol plant would assemble a drive around a Sony mechanism taken from audio DAT and HP circuit boards from Grenoble, and that a 'tenacious fight' followed between the DDS format backed by HP and a more feature-rich Data-DAT format led by GigaTape and Hitachi. It says the fourth draft of DDS went to ANSI committee X3B5 and ISO in March 1989 and that more than ten companies supported it; the Data-DAT group's format was less advanced in standardisation. The HP 35450A DDS drive was announced as a 5.25-inch drive with a built-in SCSI controller, 1.3 GB capacity, a 183 kB/s transfer rate, a 20 second average access time and an expected MTBF above 40,000 hours, with volume production expected in December 1989 (all HP vendor claims in a trade abstract; the abstract also gives prices and unit forecasts that are not reproduced). The 1.3 GB at 183 kB/s works out to about 7,100 seconds, or just under two hours (calculation), which matches HP's 'less than two hours' statement. Wikipedia and a Fujifilm seminar excerpt both give the first DDS cartridges as 60 m (1.3 GB native) and 90 m (2 GB native).

1989-2009: DDS generations. Wikipedia's generation table gives DDS-1 (1989, 60 or 90 m, 1.3 or 2 GB native), DDS-2 (1993, 120 m, 4 GB), DDS-3 (1996, 125 m, 12 GB), DDS-4 (1999, 150 m, 20 GB, also called DAT 40), DAT 72 (2003, 170 m, 36 GB) and DAT 160 (2007, 80 GB), with DAT 320 announced in November 2009 (160 GB native, 12 MB/s). The Fujifilm seminar excerpt agrees on the native capacities of DDS-1 through DAT 72 but labels the 90 m, 2 GB drives as a 'second generation' introduced in 1991 and lists DDS4 in 1999 and DAT-72 in 2003; Wikipedia gives DDS-2 as 1993, so the date of the second generation depends on whether the 90 m DDS-1 extension or DDS-2 is counted. The DDS-2 to DAT 72 dates rest on Wikipedia alone. HP's 2009 QuickSpecs confirm the DAT 40, DAT 72, DAT 160 and DAT 320 native capacities of 20, 36, 80 and 160 GB and native sustained transfer rates of 3.2, 3.2, 6.9 and 12 MB/s (HP vendor document), and say DAT 160 and DAT 320 use 8 mm wide helical-scan tape while earlier generations use 4 mm. Wikipedia says DDS-3 introduced PRML (see the PRML read channel card); the QuickSpecs list PRML among the encoding methods and LDPC for DAT 320 media, without saying which generation introduced which.

How the DDS recording works, per HP. The 2009 QuickSpecs describe a helical-scan format where data is written diagonally across the tape, with adjacent tracks at different azimuth angles and overlapping so that no guard bands are needed; the drum is tilted about 6 degrees from vertical and has four heads (two write and two read) at 90 degree intervals, so each rotation writes two tracks, and the read heads verify what was written and trigger a rewrite when needed. Compression is the DCLZ Lempel-Ziv algorithm and error correction is Reed-Solomon; the same document states a read error rate of 1 x 10 to the minus 17 bits for DAT 40 to DAT 320 and an MTBF of 125,000 hours at 100 percent duty cycle (vendor claims, not tested here).

Why it ended where it did. Wikipedia says that DDS competed with LTO, AIT, VXA and Travan, that AIT, Travan and VXA are no longer mainstream, and that LTO capacity has far exceeded the last DDS standard. The sources read for this card do not give an end-of-sales date or a dated statement from a manufacturer withdrawing DDS, and Wikipedia's note that a Gen 8 DAT was cancelled carries a citation-needed tag in the copy read, so no end date is stated here.

What the sources do not settle. The date of the second DDS generation (1991 or 1993); the sustained capacity of a Datasette side (about 48 KB, about 100 kB, or turbo-tape figures between 300 and 1000 kB); the exact effective rates of the 1980s machines, for which only a hobbyist Q&A and GitHub notes were read; whether the 3.81 mm width came from the CBS proposal or was Philips's own choice, which rests only on a search excerpt of a 2013 interview; the end of DDS production; and whether the DAT-for-audio copy flag changed the market for DAT recorders, which this card does not claim.