The 9830 was under development in the very early 1970s. Not surprisingly for the period, it is implemented primarily with 7400-series MSI and SSI TTL ICs, along with some bipolar ROMS for microcode, and early MOS ROM and RAM memory chips. It was a little too early for use of 74LS devices, but there are some 74H, 74L and 74S series ICs (high-speed, low-power and Schottky series). 512*8-bit MOS ROMs are used to store the firmware and 1K*1 1103 MOS dynamic RAM chips for read/write memory.

There are 77 different types of IC used. With 8KW of RAM there are a total of over 460 ICs.

As was common for Hewlett-Packard in that era, the printed circuit boards are gold-plated over all the copper, not just on the edge-connector fingers. Most of the boards are double-sided, although three are multi-layer: the LED display, RAM memory and M-register boards.

The full schematic may be referenced for greater detail relating to the following descriptions.

Processor

While the 9830 CPU is based on the architecture of the earlier HP 2116 series, the hardware implementation is entirely different. The 2116 was a full 16-bit parallel implementation, the 9830 is a bit-serial microcoded design.

CLOCK:

The master clock (MCK) is an 8 MHz crystal oscillator. Several clock signals are derived from this, the two primary ones being the shift clock (SCK) and the ROM clock (RCK). The shift clock is the bit-stream clock for the registers and dataflow. The ROM clock is the microcode-cycle clock, in each microcode cycle one microcode ROM retrieval and micro-operation is performed. The diagram to the right shows a full 16-bit microcode cycle. Many microcode operations do not need to process a full 16-bit word though, so a nice trick is played with the microcode cycle allowing it to vary in length (number of bits or shift-clock cycles). 4 bits (CC3,2,1,0) of each micro-operation specify how many bits long the cycle shall be, from 1 (CC=0) to 16 (CC=15).

The number of bits processed in a micro-operation can be further affected by the microcode IQN bit and selected condition. This does not affect the time period of the micro-cycle however, it just suppresses the shift clock.

REGISTERS & DATAFLOW:

The R, S and T busses are the main bit-stream data busses, R and S taking output from selected registers to become the operands for the ALU, while T takes the output from the ALU to selectively feed into registers. The P register is the program counter. The Q register, referred to as the Qualifier register in the patent, receives the program machine instruction to execute. A and B are the two accumulators or general-purpose registers. The M and T registers are used for the memory address and memory data transfer respectively. The E register is used for a carry flag and is involved in some of the BCD instructions.

In general, during micro-operations, the specified source registers will be rotated, with a data bit-stream coming out onto the R or S-bus LSB-first. The destination register receives the bit-stream MSB-first. For a full 16-bit-shift operation the source register will return to its original state by the end of the operation. For operations with less than 16 shifts (and more than 0) the source register will be left altered by the partial rotation.

ALU:

The ALU is very simple in construction, taking advantage of the bit-serial architecture to be implemented with two 256*4-bit bipolar ROMS. One of the ROMS forms the binary ALU: two operand bit-streams, the output of the binary-carry flag/flip-flop, and a 3-bit function code form 6 of the 8 bits of address for the ROM. The ROM contents of course hold the various 1-bit function-result values. One of the data outputs becomes the function-result bit-stream (T-bus) and another the input for the binary-carry flip-flop (BC).

The ALU also implements 4-bit-parallel BCD-nibble add and subtract. A microcode control line further addresses the binary ALU, switching it to another dataset in its contents. The outputs together with 3 more bits of each operand, taken from the A and T registers, address the second 256*4 ROM - the BCD extension ROM. The additional outputs form the other 3 bits of the BCD nibble result and the input for a BCD-carry flip-flop (DC).

The ALU functions are detailed further in the ALU ROM Functional Listing.

Note the binary carry flag is only seen at the microcode level, not at the machine-instruction level. The carry resulting from an add instruction is saved in the E-register for use at the machine-instruction level.

MICROCODE:

The microcode store is 256 words * 28 bits, implemented in seven 256*4-bit bipolar ROMS. An 8-bit register holds the ROM microcode address, distinguished as the 4-bit primary (PA) and 4-bit secondary (SA) portions. 8 bits of the microcode (PM & SM) will modify PA and SA to determine the next address. The address register, built up from individual JK flip-flops, is set up so the PM and SM modifier bits will or will not toggle the corresponding PA and SA address bit. Effectively, the current address and the modifier bits are XOR'd to become the new address.

The microcode fields may be broadly grouped as:
A few special functions are derived from otherwise meaningless combinations of field values, such as the QPM function which results if both 1 and A/B are fed to the R-bus at the same time.

The Q register contents may be modified by the microcode in the course of execution of an instruction, so references to Q by the microcode are not always to the 'proper' instruction encoding. A multi-way branch is performed by directing several bits of the Q register to become the primary-address modifier.

Two-way branches are implemented by selecting 1 of 16 1-bit conditions, the state of the condition will affect the lowest bit of the microcode address (SM0). The ROM PM bits do double-duty as the branch condition selector, consequentially when a branch decision is specified (BRC), the next primary-address bits are fixed to the XOR combination of the current address and the branch condition selection. With this constraint on specifying the next address, the microcode program tends to jump all over the ROM address space.

A flowchart of the microcode is presented in the patent. The ROM data contents were included in the patent and have been used to produce a disassembly of the microcode. The following tables are a summary of the micro-operation fields and encoding:

Microcode Operation Fields & Encoding
Micro-operation Fields
 PM    SM   BRC   CC   IQN  XTR  RC   SC  AC   XC   TTM  TTT
3210  3210   *   3210   *    *   210  10  210  210   *    *

PM3,2,1,0 = primary-address modifier
SM3,2,1,0 = secondary-address modifier
BRC       = branching enable, PM will select condition
CC3,2,1,0 = length of micro-cycle (CC+1 bits)
IQN       = suppress SCK if not condition
XTR       = A/B-reg to R-bus
RC2,1,0   = R-bus source selector
SC1,0     = S-bus source selector
AC2,1,0   = ALU function code
XC2,1,0   = T-bus destination & misc. selector
TTM       = T-bus to M-reg
TTT       = T-bus to T-reg
Condition Selection
PM
 0  Q0  = Q-reg bit 0
 1  Q1  = etc
 2  Q2  =
 3  Q3  =
 4  Q4  =
 5  Q5  =
 6  Q6  =
 7  BC  = binary carry
 8  P0  = P-reg bit 0
 9  Q15 = Q-reg bit 15, for direct/indirect
10  QMR = memory reference (Q*)
11  Q10 =
12  QNR = interrupt
13  Q8  =
14  DC  = BCD carry
15  QRD = I/O state machine active
R-bus Selector
RC
0  UTR = 1 -> R-bus
1  PTR = P-reg -> R-bus
2  TRE = T-Reg -> E -> R-bus
3  WTM = write T-Reg -> Memory(M), 0 -> R-bus
4  TQ6 = T-bus -> Q6, 0 -> R-bus
5  QTR = Q-reg -> R-bus
6  RDM = read Memory(M) -> T-reg, 0 -> R-bus
7  ZTR = 0 -> R-bus
S-bus Selector
SC
0  ZTS = 0 -> S-bus
1  MTS = M-reg -> S-bus
2  TTS = T-reg -> S-bus
3  UTS = 1 -> S-bus
X Selector
XC
0  TTQ = T-bus -> Q-reg
1  QAB = select A/B register from Q10
2  BCD = BCD mode for ALU 
3  TBE = T-Bus -> E-reg -> R-bus
4  CAB = flip A/B selection
5  TTP = T-bus -> P-reg
6  TTX = T-bus -> A/B-reg
7  NOP = no op
ALU Function
AC
0  XOR     = Exclusive OR
1  IOR     = Inclusive OR
2  AND     = AND
3  ZTT     = 0 -> T-bus
4  ZTT.CBC = 0 -> T-bus,   0 -> binary carry
5  IOR.CBC = Inclusive OR, 0 -> binary carry
6  IOR.SBC = Inclusive OR, 1 -> binary carry
7  ADD     = 0 -> R-bus
Derived Functions
XTR*UTR = QPM = primary address modifier comes from Q (Q14,Q13,Q12,Q11*Q14)
XTR*PTR = IOS = start I/O operation if Q10=1
          * set single-service FF via nSRA if Q10=0

BCD*QAB         = 0 -> DC                    \ clear BCD carry
BCD*UTR         = 1 -> DC                    \ set BCD carry
BCD*nUTR*(AC=3) = A3..0 +    T3..0  -> A3..0 \ BCD add
BCD*nUTR*(AC=7) = A3..0 + 9-(T3..0) -> A3..0 \ BCD add 9's-complement

Instructions require multiple micro-operations, the number varying with the type of instruction. The implementation appears to have prioritised hardware minimisation over speed, even simple instructions tend to require quite a few microcode operations. The ADA/B instruction (add memory to register), for example, involves 12 micro-operations. Below is the typical sequence for the ADA/B instruction:

Example Micro-operation Sequence: ADA/B Instruction
Address    Next Address   Shifting Special           Dataflow -->               Description
--------  --------------  --------  ---  ----------------------------------- ---------------------------------------------------------------------
                Br  Fail  SCK  Sh                                             
Dec PASA  PASA Cond PASA  Cyc Cond       XTR   RC    SC    AC    XC  TTM TTT
=== ====  ==== ==== ====  === ====  ===  === ======= === ======= === === === =====================================================================
234 1612  0616             12                RDM.ZTR ZTS ZTT.CBC NOP         load T with instruction from memory
110 0616  1213 QNR  1212   16                  UTR   TTS   AND   TTQ         transfer T to Q, also branch for interrupt

171 1213  0012 QMR  0013    1                  ZTR   ZTS   ZTT   QAB         select A/B, branch for memory-reference class

 10 0012  1313 Q10  1312   10                  ZTR   TTS   IOR   NOP TTM     load M with 10 LSBs of address from instr, branch for base/current page
187 1313  1102              6                  ZTR   MTS   ZTT   NOP         rotate M 6 more bits to form current-page address
146 1102  0003 Q15  0002   12                RDM.ZTR ZTS ZTT.CBC NOP         load T with operand from memory, branch for direct/indirect

  2    2  1700              1  QRD  QPM  XTR   UTR   ZTS   ZTT   NOP         multi-way branch on instruction type

  0    0  0005             16            XTR         TTS   ADD   TTX         Add: A/B + T -> A/B
  5 0005  0704  BC  0705    1   BC             ZTR   UTS   IOR   TBE         set E bit if binary carry set, and branch
116 0704  0202              3                  ZTR   ZTS ZTT.CBC TBE         clear 3 more bits of E if binary carry was set

 34 0202  1603              1                  PTR   UTS   ADD   TTP TTM     add 1 to P at LSB, load result into M
227 1603  1612             15                  PTR   ZTS   ADD   TTP TTM     ripple-carry through 15 bits of P, load M
                          ---
                           94

INSTRUCTION EXECUTION RATE:

At 8 MHz, the raw master clock rate sounds impressive for the era. The bit-serial implementation quickly reduces that: at 16 bits and some overhead the basic processing rate for a full word is reduced to less than 500KHz. The variable length of micro-operations gets some of the speed back. There is a 1 MCK-cycle overhead for each microcode cycle, so in practice: 
microcode cycle = (1 + (1..16)) * 125nS
                = 250 .. 2125 nS
                == 4MHz .. 471KHz

The ADA/B instruction listed previously takes 94 MCK cycles, with 12 more for the per-operation overhead: 

ADA/B time = (94+12) MCK * 125nS/MCK
           = 13.25 µS

This is the equivalent of a not-very-speedy 75,000 instructions per second. In comparison, on the HP 2116 an ADA/B instruction takes 3.2 µS == 312,000 IPS.

RAM memory refresh further reduces the effective instruction rate. As explained further below, the processor is halted for 20 out of every 840 µS for refresh, reducing the processing rate by 2.4%.

Memory

The architecture provides for a 32 KW address space. The lower 16KW is reserved for the firmware ROMs. Varying amounts of RAM were in the upper 16KW.

ROM:

The ROM chips are 512 * 8 bits, and were produced by HP. For some reason the 8'th address bit is not decoded internally in the ROMS and instead comes out as 2 pins, 1 each to enable 2 banks of 256 bytes. As a result, there ends up being extra logic and an extra address enable line (both A8 and nA8) trailing around the circuitry.

7.5 KW of ROM are standard/required for the firmware, split between between the BASIC ROM board (7KW in 28 chips) and the BASIC II ROM board (512W in 2 chips, 512W of unused address space). The other 8KW space addresses the option ROMs (3 slots, 1KW each) and cartridge ROMs (5 slots, 1KW each).

RAM:

There were 3 versions of RAM board-sets produced over the production lifetime of the 9830. The "original" set for the 9830A provided 2KW per board. According to the 9830 Service Manual, the original set was installed in only the first 200-or-so machines.

The original set was followed by the "standard" set, which had both 4KW boards and 2KW boards, but also new M-Register and T-Register boards.

Both the original and standard set were implemented with the historic Intel 1103 1K*1 dynamic RAM chip, the first RAM IC to start putting a dent in the hegemony of core memory. The 4KW boards were densely packed with 64 1103 chips.

There are slots in the backplane for only two memory boards, so the machine was limited to 4KW with the original set and 8KW with the standard set. Note that the architecture allows for a maximum of 16KW RAM. Some comments in the patent and extra signals on the external I/O slots suggest there may have been the intention to provide for expanding RAM memory in an external case. The extra signals - which were not needed by general I/O interfaces - were used by the disk I/O sub-system.

The 9830A was superseded in 1976 by the 9830B, the primary difference being the ability to increase the RAM from 8KW to a full 16KW. This was accomplished with a third RAM board-set, with 8KW per memory board using TMS4060 4K-bit chips. Once again, a new M-Register board was also required.

HP9830 RAM Board-Sets
Board-Set M-Register Board T-Register Board Memory Board Chip Chip Size Words / Board Max. RAM Memory
9830A Original 09830-69522 09830-69523 09830-69524 1103 1K * 1 2 KW 4 KW
9830A Standard 09830-66582 09830-66583 09830-66584 1103 1K * 1 2 KW 4 KW
9830A Standard " " 11275-66584 1103 1K * 1 4 KW 8 KW
9830B 09830-66592 " 09830-66594 TMS4060 4K * 1 8 KW 16 KW

(See HP-9830A Service Manual / Jan 76, Chapter 6. Thanks to Mattis Lind for the 9830B information.)

The "standard" board-set was the most prevalent and the remainder of information on this site is derived from and focussed on this set.

RAM REFRESH:

Refreshing of the dynamic RAM chips is accomplished by simply halting the processor periodically and executing a sequence of memory access cycles while halted. In each refresh period, 32 memory access cycles are performed, one for each row in the 1103 chips. The refresh memory access cycles each require 5 MCK cycles. An additional MCK cycle is needed to transition into the refresh period, thus: 

refresh period = (32*5 + 1) * 125nS
               = 20.125 µS

The run period is determined by the refresh delay monostable, and so varies with RC tolerances. The calculated design target for the monostable is 820 µS (based on the RC values on one instance of the M board). The start of refresh has to wait for the end of the current microcode cycle, so the run period may be up to ~ 2µS longer. Each refresh cycle only refreshes half the RAM memory banks, with a given bank being refreshed on alternate cycles. The refresh rate is approximately: 

refresh rate = 2 * (820+2+20)µS
             = 1684 µS
             == 594 Hz

The 1103 datasheets specify a minimum refresh rate of 500Hz (2mS).

MEMORY CYCLE STATE MACHINE:

The access cycles for the 1103s are complex enough that a small 4-bit state machine is used to sequence the control signals for the 1103s. The memory cycle state machine implements 3 types of cycle: read, write and refresh. Some analysis and a spreadsheet simulation of the state machine logic indicates a read-cycle takes 10 MCK cycles, a write-cycle takes 11, and - as mentioned above - a refresh-cycle takes 5. The microcode flowchart indicates 10-bit microcode cycles for both read and write, while the patent microcode ROM data specifies 12-bit cycles for both read and write.

The same state machine and read-cycle are used for reading from the ROMs even though the accessing requirements are much simpler. It would be interesting to know what the speed specifications for the HP ROM chips actually was, in that if it was significantly faster than the > 1 µS cycle time of the state machine, it should have been possible to implement a 4th and faster cycle for ROM reads.

ROM / RAM SWAPPING:

The processor expects RAM at some reserved locations due to some addresses set in the microcode. The architecture also breaks memory into pages at the instruction level and accessing within the zero-page (and the current-page) can be more efficient than an arbitrary memory access.

For such reasons, a small amount of RAM and ROM are swapped so there will be some RAM within the zero-page (lowest 1KW), for use by the system. Some unfortunate logic in the address decoding swaps 256 bytes of ROM at the top quarter of the zero-page (001400-001777) with the first 256 bytes of RAM (040000-040377).

Input / Output

The 2116 architecture set out a fairly straightforward and consistent I/O architecture, with a 16-bit I/O bus and each device having a control-bit and a flag-bit (more detail). There were 9 I/O instructions and devices were addressed by a channel number embedded in I/O instructions.

At the machine-instruction level, this scheme has been retained in the 9830, but another level of complexity has been added.

SELECT-CODE DEVICES:

The new scheme may be described as having sub-devices hanging off channel 0 and 1. The devices having a 0..15 select code as seen at the user level are all in this new class of sub-devices. They are referred to here as Select-Code devices, to distinguish them from Channel devices. The tape drive, the 9866 printer interface and any device plugged into the external I/O bus are all select-code devices.

A select-code device is addressed by specifying the select-code in the upper 4 bits of a word output to channel 1. Usage of the other 12 output bits is determined by the specific device but general policy is the lowest 8 bits are a byte of data and the remaining 4 bits a command/state directive. A similar policy applies for input words, the lowest 8 bits are data, the next 4 are device state information. The uppermost 4 are not used on input.

CHANNEL DEVICES:

The 2116 provided 6 bits for the channel number in the instruction encoding, in the 9830 this is reduced to 5 bits. Most of the internal devices are channel devices. With a limited number of them, the hardware does not bother to fully decode the channel bits but addresses devices with one bit per device. Channel number 0 is decoded, allowing for 6 channel devices: 0, 1, 2, 4, 8 and 16. As mentioned, channels 0 and 1 are dedicated to the select-code devices, the display is at channel 8, and the beeper is triggerred by a STC instruction (set control bit) on channel 2. The keyboard occupies both channel 16 and select code 12. Channels 2 and 4 were dedicated to the magnetic card reader and internal printer in the 9810/20 calculators. There are a few gates on the CPU board 11 used for these channels, superfluous for the 9830.

I/O operations and devices are covered in further depth on the Machine Architecture page.

I/O STATE MACHINE:

In the hardware, there is another state machine to handle the I/O instructions. The processor hands off the instruction to this state machine and waits for it to indicate completion (via the QRD flag). Note the processor is only waiting here for the instruction execution, a relatively quick operation, not for the device to complete some command.

Alphanumeric LED Display

The 32-character display is made up from eight 4-character 5*7 dot-matrix LED displays. These were made (and sold) by HP.

For multiplexing, the display is broken into 2 groups of 16 (lower and upper halves of the display line) and the two groups scanned in parallel, column by column. For example, character positions 3 and 19 are scanned at the same time, and at a given instant there may be up to 14 LEDs being illuminated. The duty cycle per LED or column is then ~ 1/80=1/(16*5). The on-time for a column is 88µS, determined by two monostables forming a gated oscillator. A digit is scanned in 440µS, and the full display in ~ 7mS.

The LEDs are physically organised into a 28-row * 40-column matrix, not 14*80 as might be anticipated (or 7*160). A total of 68 drivers are then required, which is as close as could be expected to the 2D minimum of 67 = 2*SQRT(5*7*32).

The display hardware takes care of generating the 7-row pixel patterns with a Texas Instruments MOS character-generator ROM, and of scanning the 5 columns within each character. The firmware is still involved in the multiplexing however. The firmware must periodically deliver two 6-bit ASCII character codes to the device along with a 4-bit address to select the character positions, via the I/O Data Output bus (DOx), and trigger the display-enable (DEN) signal. DEN enables the gated oscillator for the 5-column scan of the two character positions. The oscillator multiplexes the character ROM between the two character groups - note the 7-bit latch for one group at the output of the ROM. The other group relies on stable data on the DO bus through the ROM during the scanning interval.

The gated oscillator is formed by the two monostables of IC U42-c3, the other two monostables in the circuit (U42-c2) are not strictly necessary functionally, they appear to be present to protect the LEDs in the event of failure in the firmware scanning. In particular, the scanning could be upset by a problem on the DO bus which is externally accessible via the I/O slots.

The design and the ROM character generator work on the opposite scanning orientation to the character generators typically used for raster-scan video terminals. The ROM is from the TMS4100 series of character generators produced by TI, but (as Rob pointed out) the lazy-T prompt character of the 9830 is not a standard character, so the ROM was not an off-the-shelf IC.

At the machine level, the display device is located at I/O channel 8. An "OTA/B 8" instruction sends the ASCII codes and character position to the device and sets the DEN flag. A "CLF 8" instruction clears the DEN flag.

Tape Drive

The tape drive uses cassettes of the same form factor as standard audio cassettes, however it records data in native digital form. It does not use an audio modulation scheme such as FSK. Nor does the drive use a typical audio cassette mechanism - it does not use a capstan to move the tape. The tape reels are driven by a motor each and the driven-reel RPM is constant during an operation, so the tape velocity varies with the amount of tape on the reel due to the changing diameter. The drive operates at three speeds, fast for rewinding and fast searching, normal for writing and reading, and slow when drive movement begins in leader.

BIT & BYTE ENCODING:

The tape drive receives and delivers parallel-bit bytes at the interface to the drive. The drive performs the byte-to-on-tape-bit-stream conversions.

Two channels on tape are used to encode a single bit stream. A flux pulse in one channel indicates a zero in the bit stream, a flux pulse in the other channel indicates a one. This scheme makes clock extraction trivial and is tolerant of wide speed variation.

Flux pulses occurring simultaneously in both channels are a mark and are used as a byte delimiter. There are nine content bits in a byte: 8 data bits and a control bit. With the mark bit there are an average of 10 bits/byte. When writing, a mark is written at the beginning of the first byte and at the end of every byte, so there is a mark at the beginning and end of every byte, but only one mark between bytes. When reading, bytes must be fully formed - 9 bits between marks - to be recognised by the drive as a valid byte. This avoids delivery of incomplete bytes when reading begins in the middle of a byte, or bytes which were malformed by partial overwrite when a portion of a tape is rewritten.

The control bit is used to indicate a control - or interrupting - byte. Only one control byte is defined: the Beginning-of-File byte (control-0x3C). When finding a file, the drive is set to read in control mode, and (only) BOF bytes are recognised by the drive, and generate an interrupt to the processor. Note also that 0x3C, with the control bit in the middle, has a symmetric bit order, so the same hardware will detect the BOF byte when reading in reverse as well as forward.

When writing, head current in both channels is turned on at the beginning of the write. The bit period is determined by two monostables - one of ~ 200 µS and one of ~ 130 µS - configured as a loopback oscillator. To produce a flux pulse, the current in a channel head is reversed for the 130 µS interval. As current is always flowing in the head in one direction or other during writing, any prior data on the tape is erased. The basic bit and byte rates are: 

bit rate = 200µS + 130µS
         = 330 µS/bit
         == 3000 bits/S

byte rate = 330 µS/bit * (8+1+1) bits/byte
          = 3.3 mS/byte
          == 300 bytes/S

Anomaly: The drive may read in either control mode or data mode. In control mode, normal indication of data bytes is inhibited and only control bytes will be indicated at the interface. In contrast, in data mode all bytes are delivered to the interface but without the control bit. In data mode, there is no way to distinguish a control byte from a data byte.

FILE & TAPE ENCODING:

A tape contains a sequence of files separated by zeroed padding bytes. Each file consists of a Beginning-Of-File control byte, a header area and content area. The header and content areas have separate checksums.

Aside from the padding bytes and the BOF byte, all data on the tape is in terms of two-byte 16-bit words, with the low-order byte coming first. The header fields are: 

FileID         = 0..n sequence number of file
FileLength     = number of content words in use by file, not including checksum
FileType       = file type
FileSpace      = number of words for content reserved on tape, i.e. maximum FileLength
FirstLine      = first line number of program
LastLine       = last line number of program
CommonAreaLen  = number of words of program common area
Checksum       = sum of preceding header or content words, modulo 2^16

FileType   = 0 : file not in use
             1 : Binary block
             2 : Data
             3 : BASIC program
             4 : Key macros
             5 : Special program file
            21 : Secured binary block
            23 : Secured BASIC program
            24 : Secured key macros
            25 : Secured special program file

There is no explicit last file or end-of-files indicator.

TAPE TRACKS & HEAD AMPS:

Each channel is half the width of the tape, the recording is thus 2-track and single-sided. Flipping a tape around may overwrite data on the 'other side'.

Note that normal stereo audio cassette recording is 4-track and double-sided (2 sides, 2 tracks per side). Normal mono audio cassette recording is 2-track and double-sided (2 sides, 1 track per side). As such, neither of the common cassette heads is compatible with the 9830/65 head.

By calculation from component values, the gain for the head amps is set in the range of 500 to 600. The output from the head amps is typically around 1.5 to 2V (observed), the output from the heads should thus be around 2.5 to 3mV.

OPTICAL SENSOR:

The optical sensor in the drives is used to sense the Beginning and End of Tape. It uses an incandescent grain-of-wheat lamp (58mA@5V, 64mA@6V, 0.25inL*0.125inD) emitter and a Light-Dependant Resistor (LDR) sensor to distinguish between clear leader tape and opaque magnetic tape. A good LDR will display a resistance of several mega-ohms when fully dark and less than 200 ohms under strong direct light. The LDR feeds into a 741 op-amp running open-loop, thus functioning as a comparator. The comparator output will trip when the LDR resistance is around 18K.

The sensor relies on the cassette having a transparent leader for optical transmission and a sufficiently reflective case to reflect light from the lamp back to the LDR. If the cassette case is not sufficiently reflective, if the tape leader does not have sufficient transparency, or if the bulb burns out, the drive will not be able to detect the BOT/EOT and not know to stop the motor at end of tape or when rewinding.

The LDR sensors can also fail. One failure mode is the LDR becomes leaky and will not reach a high enough dark resistance to trip the comparator. In one instance, a faulty LDR would not go above ~ 14K when dark, which is too low for the comparator to recognise as dark. In such cases the drive will always think the tape is in leader, hence at the BOT or EOT. The drive will run when commanded but run at the 'slow' speed because it thinks it is in leader. Pressing the REWIND button will rewind tape very slowly while the button is held, but the drive will not latch into rewind.

In operation, the LDR sense level and comparator input can be observed at A65-TP1. When magnetic tape is over the sensor, TP1 should be less than 2.5V. For leader, TP1 should be greater than 2.5V. As measured in one unit with the original lamp/LDR pair: tape=0.1V, leader=4.5V. These imply resistances for the LDR of tape=900K, leader=2K.

It is possible to replace the incandescent lamp with an LED and the LDR with a phototransistor.

Anomaly: A transition into leader at either BOT or EOT will stop the drive. When tape movement is initiated when in leader, the drive begins at slow speed but must continue regardless of the direction specified as it cannot distinguish between BOT and EOT. So, for example, a rewind command when the tape is already rewound will result in the drive jamming on attempting to rewind the tape, and similarly for a forward command issued when at EOT.

CASSETTE TAPES:

HP sold special cassettes for use with the 9830 and 9865, to meet the particular requirements of the drives mentioned above:

In practice, ordinary audio cassettes can be used, as long as the leader is transparent. Non-reflective cases may be modified by painting the appropriate surface in the cassette white or sticking a piece of white paper at the back of the compartment.

The actual magnetic tape does not seem to be critical. The recording method is a fairly simple and low density saturation recording. Cheap audio cassettes and CrO2 types have been successfully tried.

Pressure Pad Problems: Cassettes have an integral pressure pad behind the tape to hold the tape in close contact with the head. Two forms of pressure pads have been observed: a piece of foam with felt pad, and a metal band spring with felt pad. This applies to both audio cassettes and the HP-marked ones.

In many cassettes with the foam pad the foam has deteriorated, resulting in poor signal levels and erratic read/write operations. The foam pressure pads can be rebuilt with some new foam and double-sided tape, or in some cases a band-spring pad may be scavenged from another cassette and used to replace the foam-type pad.


  Gallery | Machine | Hardware | Microcode | ALU | Memory Cycles | Unit Log
HP 9830 		bhilpert
2013 Oct