Epson HX-20 laptop

The Epson HX-20 is a very early laptop from about 1982. It's got a pair of H63C01 CPUs running at 614kHz, a microcassette drive (which can be swapped for a ROM cartridge), a printer, and a built in 4 line graphics LCD, and 16kB of RAM. Mine comes with the expansion unit which gives it an extra 16kB of RAM. It's got an internal OS in ROM and a pretty good BASIC in ROM as well. There's space internally for an extra ROM and the expansion unit has room for two more.

One nice feature is that it could run for quite a while off internal NiCD batteries, and they would preserve the RAM while it was off; together with the OS in ROM, it meant it was ready to use instantly.

In some ways, it's as much an exhibition of Epson's capabilities with the tiny printer, they sell similar ones for tills, the LCD and the microcassette. A good stash of documents is on Phil's page. The documentation is excellent, although the Tech reference manual doesn't always agree with the main schematic, which in general is correct.

1. Fixing the hardware
2. The Tape
3. Intext
4. TODO

Fixing the hardware

As I received it, it was clean and basically worked. The NiCD's had been removed and a power lead bodged on the NiCD connector. It ran on 6v but the LCD was faint, and I had problems with stability and with the RS232 port. Also, it really shouldn't need 6v, the NiCD was supposed to be 4.8v max, with a 4.5v low voltage cutoff.

It's pretty easy to work on, the two halves of the case fold out (after unplugging a couple of ribbon cables) and the main board is a simple two sided PCB with through-hole components. The back of the main board is very well labelled, so finding pins to check is easy. One constraint is that there's not much height, so they used pretty flat components. After some basic checks I started digging further.

LCD caps

RS-232

One mistake I made was concentrating too much on short spikes on the power caused by the DC-DC from the 232. Setting a trigger on the 4.5v (which is what triggers the 'charge battery') and zooming in on the spikes initially got me hideous output like:

The ~40kHz repetition is the frequency of the DC-DC, and we can see a truly horrid drop to under 3V and a peak of 6.52v (higher than the input supply). While this is awful, it hides a deeper problem which is only seen on a slower timebase (albeit taken closer to the DC-DC than the last one): That slow drop off made me realise the problem was not a high frequency issue; tracing that back led me to the bodged battery in connector which was high resistence, but didn't really cause a problem until the higher current load when the 232 was operating. Replacing the battery lead got me the, slightly more sensible: While this is still very spikey, it's worth noting it's on the Q2 transistor right next to the DC-DC. I finally added an extra 1uF capacitor between R2 and the nearest ground I could find (which is still a little bit far away).

With all these fixes, I could run some basic 232 testing right down to 4.9v input.

Batteries - or lack there-of

Now that I was running with the capacitors in the battery box, I was supplying power via the original input socket. I noticed that the diode D2 (a S5277B), has a drop of about 0.7v, I replaced this with a 11DQ10 whose drop is closer to about 0.5v. This lets me keep the capacitors full at about 5.8v input, and even 5v input would be useful to keep memory contents. Note: It's not safe to remove the diode, it's not only for reverse input protection, it also stops the batteries feeding back to the power input.

Power plug - warning

Status

The Tape

Examining the listing it's a simple till program for a Garden centre in St.Albans, listing items such as Shrubs, Conifers, Bedding Plants etc. I captured the listing over 232 to Minicom on a Linux PC by using the command:

SAVE"COM0:(68N1F)",A

Intext

(A fun aside is that the RAM chips in the expansion are 6117 rather than the more common 6116. They have a lower power mode where they disconnect from the bus more to save power, this does lead to them having slightly different enable signals).

Intext is a very simple word processor and comms package. One of it's uses was for journalists to enter a story on the road and upload it (over an acoustic coupler). It has the ability to embed simple formatting controls, but isn't WYSIWYG on the display. The manual explains that they've agreed the code with the Journalist and typographical unions!

Selecting Intext gave an error about not having enough memory though and to look at the manual. Alas the only manual I could find was for the Cassette version which didn't mention the error.

My friend had assumed that this was a RAM fault, but PRINT FRE(0) from BASIC showed the magic 29275 bytes free that everyone was expecting there to be with the expansion unit. I later confirmed that I could read and write the expansion unit RAM by using the 'S' command in the monitor.

Getting to the Intext EPROM

Entering the monitor normally shows the BASIC ROM paged in, at one - non repeatable - point I managed to get Intext to trap as it printed its error out, and that landed in the monitor, and I could see the INTEXT header around 8000. Alas that was not repeatable. I eventually managed it by running the BASIC commands:

POKE &H7E,&H80
POKE &H30,&H30

It's easy to show the memory from the monitor using D 8000 and I also verified that it seemd to be 8kB, not 16kB. The monitor offers the 'W' command to write out an area of memory, alas this only allows writing to the internal or external cassette, and not to serial. Based on the documentation in the Operations manual chapter 9, I came up with the magic:

A
  T 8000
  L A000
  O 0
  E 0
WC,INTEXT.BIN

Audio to file

parecord -r --channels=1 epson.raw --file-format=raw --format=u8

Decoding the audio

Looking at the output file showed the INTEXT header near the beginning. But the file was 8960 bytes rather than the 8192 I had expected.

After some time I realised that this was because the file was in SAVEM format, and had to write another level of decoding. Thus that repo also includes desavem to extract the contents. This left me with an 8193 byte file - I think the extra byte is because the 'L' of the monitor command should be one byte before the end.

Note that neither of my programs checks the CRC or checksums.

Disassembling and understanding

git clone https://github.com/NationalSecurityAgency/ghidra.git
git fetch origin pull/6314/head:h6031

I could easily see the expected entry header, that I'd also seen in the Epson's Monitor:

            8000 ba              ??         BAh
            8001 41              ??         41h    A
            8002 ff              ??         FFh
            8003 ff              ??         FFh
            8004 80 0d           addr       entry
            8006 49 4e 54        ds         "INTEXT"
                 45 58 54

I could also see the error text I was trying to understand:

str_memtoosmall
        80d7 4d 65 6d        ds         "Memory too small.   See Manual.\f"
             6f 72 79 
             20 74 6f 

       entry
            800d 96 78           LDA        A,INIFLA_0078
            800f 84 80           AND        A,#0x80
            8011 27 47           BEQ        print_notenoughmem

OK, so the first thing then is to make sure I start BASIC before I try starting INTEXT. Doesn't help, OK, digging further I see that Ghidra is showing a separate piece of code that ends up at the 'print_notenoughmem' routine:

            8047 fc 01 34        LDD        BSWTAD_0134
            804a c3 00 06        ADD        D,#0x6
            804d fd 7f 58        STD        DAT_7f58
            8050 fe 7f 58        LDX        DAT_7f58
            8053 ee 02           LDX        0x2, IX
            8055 8c 02 00        CPX        #0x200
            8058 2e 0c           BGT        LAB_8066
print_notenoughmem

        _DAT_7f58 = (short *)(BSWTAD_0134 + 6);
        sVar1 = *(short *)(BSWTAD_0134 + 8);
        if (sVar1 != 0x200 && sVar1 + -0x200 < 0 != (SBORROW2(sVar1,0x200) != false))
        goto print_notenoughmem;

A working INTEXT

• Initialise with Ctrl-@
• Select BASIC
• use CLEAR 50,10000 to allocate a chunk of RAM file area.
• Use the Menu button and enter INTEXT

TODO - 2025/10

• Play more with a replacement battery setup.
• Wire up a floppy emulator to the Serial (not RS232) connection
• Set up printing to it's printer or access to it's tape over the RS232 connection.

mail: fromwebpage@treblig.org irc: penguin42 on libera.chat | matrix: penguin42 on matrix.org | mastodon: penguin42 on mastodon.org.uk