● refactoringbuildingsamridhlimbu.com/projects/bed-monitor · v0.1

❯ cd projects/bed-monitor

Bed monitor

uni · SIT329 · T2 2026

A prototype bed monitor for aged care. A small board reads a temperature sensor, a pressure mat and a button, blinks an LED for a fever, and calls a nurse when someone taps the mat five times in two seconds. A dashboard on a laptop shows each bed and logs everything to a database.

The Aged care bed monitor dashboard connected to /dev/cu.usbmodem1101: body temperature 19.67 C marked hypothermia, bed occupied at 1041 g and 4095 counts, a red alarm bar reading HYPOTHERMIA and HELP, NURSE CALLED with an Attended, reset alarm button, and a link log of frames 2960 to 2969 — The dashboard on the real board, 7 Sep 2026. The thermistor was in room air, so 19.67 C reads as hypothermia. HELP came from taps on the mat. The grams come from a placeholder calibration, not a scale.

FreeRTOS tasks

814

host checks passing

1,216

board frames in SQLite

sequence gaps inside a connection

Context

Task 6.1D in SIT329, Advanced Embedded Systems, at Deakin. The unit is about programming an ARM Cortex-M0+ through its registers, so nothing in the running program calls digitalWrite, analogRead or attachInterrupt. The timer, ADC and USART classes come from my own earlier tasks in the unit.

The task sheet asked for temperature every 5 minutes, a red LED blinking once a second in a fever, an alarm above 39 C or below 35 C, bed occupancy from a force sensor, a HELP alarm on five taps in two seconds that a nurse can reset, and a dashboard that logs to a database.

How it fits together

Both analogue sensors share one ADC, so a mutex guards it and only the temperature task ever moves the multiplexer. The button interrupt stamps each press and drops the time into a queue for the mat task. Once a second the link task builds a text frame with a checksum. Each byte goes out on D6, comes back on D7 through a jumper wire, and is then passed on over USB. The dashboard checks the checksum, stores the frame and can send commands back.

Diagram: thermistor on A0 and FSR mat on D4 feed the ADC, the tap button on D2 feeds the EIC; TC3 drives the LED. The ADC and EIC feed four FreeRTOS tasks guarded by three mutexes. vLinkTask sends one frame a second through SERCOM0 and USB serial to bed_link.py on the laptop, which writes to SQLite and drives the Tk dashboard. Commands go back the other way. — Diagram of the system, with the pins and task names from the code.

Tasks and timing

Four tasks, each woken on a fixed period with vTaskDelayUntil so a late run does not push the next one back. The LED is not a task. A hardware timer interrupt toggles it, so nothing the scheduler does can slow the blink.

Timing diagram over one second: TC3_Handler fires at 500 and 1000 ms, vMatTask at priority 5 is released 100 times, vCommandTask at priority 4 twenty times, vLinkTask at priority 3 once, the timer service task at priority 2 is not used, and vTempTask at priority 1 runs every 5 minutes. — When each task is released in one second, with the priorities and periods from Task6_1D.ino.

The board prints its own setup at boot. Here is part of that banner from 24 Aug 2026. The register values and the baud rate are read back off the hardware, not copied from the source.

link
  SERCOM0 CTRLA = 0x40200106, CTRLB = 0x30000, BAUD = 63019
  asked for 115200 baud, got 115219, error 165 ppm against a limit of 20000 ppm
  D6 to D7 loopback: up
pins
  A0 thermistor, D4 mat, D6 TxD, D7 RxD, D8 red LED, D2 tap
scheduler
  tick 1000 Hz, priorities 0 to 8, heap free 13912 of 14336
  mat 5 at 10 ms, command 4 at 50 ms, link 3 at 1 s, temp 1 at 5 min

Serial output from the board, 24 Aug 2026. Lines copied as captured, trimmed to the link, pins and scheduler.

Telling a tap from a person

The same mat has to report occupancy and catch a call for help. Occupancy only changes after a reading has stayed past its threshold for 3 seconds: on above 400 g, off below 250 g. A tap has to rise and fall back within 400 ms, or it is not a tap.

Schematic of three signals: a short pulse that falls back inside 400 ms is counted as a tap; a step that stays up past 400 ms is thrown away; five pulses inside 2 seconds latch the HELP alarm. — Schematic of the rule, not measured data.

Here is the rule on the real board. The dashboard stored every frame from this session in SQLite. The mat was held down twice, then tapped quickly.

Mat force

Occupancy turns on 3 s after the load passes 400 g, off 3 s after it drops

view data

seconds	force
0	0 g
1	0 g
2	0 g
3	0 g
4	0 g
5	0 g
6	0 g
7	0 g
8	0 g
9	0 g
10	0 g
11	0 g
12	0 g
13	319 g
14	804 g
15	985 g
16	1,029 g
17	1,038 g
18	1,041 g
19	1,041 g
20	1,041 g
21	1,041 g
22	1,041 g
23	1,041 g
24	1,041 g
25	1,041 g
26	1,041 g
27	1,041 g
28	1,041 g
29	1,041 g
30	1,041 g
31	121 g
32	17 g
33	0 g
34	0 g
35	0 g
36	10 g
37	206 g
38	33 g
39	520 g
40	894 g
41	1,008 g
42	1,034 g
43	1,040 g
44	1,041 g
45	1,041 g
46	1,041 g
47	1,041 g
48	1,041 g
49	1,041 g
50	1,041 g
51	1,041 g
52	1,041 g
53	1,041 g
54	1,041 g
55	1,041 g
56	1,041 g
57	1,041 g
58	1,041 g
59	212 g
60	208 g
61	47 g
62	5 g
63	0 g
64	0 g
65	0 g
66	0 g
67	124 g
68	124 g
69	18 g
70	0 g
71	0 g
72	0 g
73	0 g
74	0 g
75	0 g
76	0 g
77	0 g
78	0 g
79	0 g
80	0 g

Taps counted

None added while held; HELP latches at 14 (68 s), reset at 74 s

view data

seconds	taps
0	0
1	0
2	0
3	0
4	0
5	0
6	0
7	0
8	0
9	0
10	0
11	0
12	0
13	3
14	3
15	3
16	3
17	3
18	3
19	3
20	3
21	3
22	3
23	3
24	3
25	3
26	3
27	3
28	3
29	3
30	3
31	3
32	3
33	3
34	3
35	3
36	4
37	5
38	6
39	6
40	6
41	6
42	6
43	6
44	6
45	6
46	6
47	6
48	6
49	6
50	6
51	6
52	6
53	6
54	6
55	6
56	6
57	6
58	6
59	6
60	6
61	7
62	7
63	7
64	7
65	7
66	7
67	10
68	14
69	14
70	14
71	14
72	14
73	14
74	14
75	14
76	14
77	14
78	14
79	14
80	14

Real readings from the board, 8 Sep 2026: 81 frames from the dashboard's SQLite store, one a second. Grams use the placeholder calibration, so 1041 g is its ceiling, not a weight.

The same alarm in the raw serial stream from 7 Sep. TAPS climbs and ALM goes from 02 (hypothermia) to 06 (hypothermia and HELP).

$AC,SEQ=482,UP=510000,BED=1,T=1934,TB=HYPO,MAT=21,F=0,OCC=0,ALM=02,TAPS=0,ST=27*6E
$AC,SEQ=483,UP=511000,BED=1,T=1934,TB=HYPO,MAT=4095,F=120,OCC=0,ALM=02,TAPS=1,ST=27*67
$AC,SEQ=484,UP=512000,BED=1,T=1934,TB=HYPO,MAT=4095,F=357,OCC=0,ALM=02,TAPS=4,ST=27*64
$AC,SEQ=485,UP=513000,BED=1,T=1934,TB=HYPO,MAT=4095,F=405,OCC=0,ALM=06,TAPS=8,ST=27*6C

Four consecutive frames from the board, 7 Sep 2026, as captured.

The dashboard after a reset: body temperature 20.03 C, bed EMPTY at 0 g, the alarm bar showing only HYPOTHERMIA, and the link log showing the load falling from 1041 g to 0 g, then sent RESET, board acknowledged RESET, and alarm word 0x06 to 0x02 — After pressing the reset button, 7 Sep 2026. The log shows the load going, the reset and the board's acknowledgement, and HELP clearing from the alarm word.

What the logs show

The SQLite file holds 1,216 frames from the real board, logged on three days: 24 Aug, 7 Sep and 8 Sep 2026. None has the stub bit set, so none was injected. The sequence number jumps six times, each time at a reconnect or a reboot, and never skips within a connection. Within a connection the temperature only changes at uptime 301 s and 3001 s, which is the 5-minute task running on time. The board also reported its own counters after 51 minutes up.

13|2026-09-07T04:47:01.890Z|1|stats|up 3083.0 s, TC3 ran 6166 times (2.00 a second), EIC took 3 edges and dropped 2 as bounce, busy waits adc 0 state 0 link 0, heap free 6704

A row from the events table (id, time, bed, kind, detail), 7 Sep 2026: the board's STATS reply as the dashboard stored it.

Key technical decisions

the LED on a hardware timer over a blink task

The blink has to keep a steady 1 second beat whatever else is running, so it is not a task at all. TC3 counts to 31249 at 62.5 kHz and its interrupt toggles the pin every 500 ms. No load on the scheduler can stretch it. After 51 minutes up, the board reported TC3 firing 6166 times, 2.00 a second.

rate-monotonic priorities over priority by importance

The task with the shortest period gets the highest priority: the mat at 10 ms, commands at 50 ms, the link at 1 s, temperature at 5 minutes. Priority 2 is left empty because the kernel creates its own timer service task there.

a tap is a pulse over any threshold crossing

The first tap detector counted someone getting into bed as a tap. Now a crossing only becomes a tap when the signal comes back down. One still up after 400 ms is a person, so it is thrown away.

a mutex on the link over two tasks sharing the USART

The command task can pre-empt the link task halfway through a frame. Each byte is sent and then read back off the D6 to D7 jumper before the next, so a pre-emption there would mix two frames together. A third mutex fixed it. No host check could catch this one, because it needs two tasks running.

one clock generator each over sharing generator 5

The timer and USART classes from earlier tasks both claimed clock generator 5 at different speeds. Nothing stops you rewriting a live generator, so it would have failed quietly: a 3 Hz blink, or every baud rate a third of what was asked. The USART moved to generator 6. The board reported 115219 baud against the 115200 asked for.

microvolts over whole millivolts

Rounding the thermistor voltage to whole millivolts threw away about half the temperature resolution between 35 and 39 C. Working in microvolts, every one of the 233 ADC counts in that range changes the reading. The host checks confirm all 233.

Testing without the board

The monitor logic is plain C with no hardware calls, so it compiles on a laptop and runs against 184 checks. The register classes run against mock registers: 630 more checks. A simulator built from the same C writes 2,761 frames over 46 minutes of virtual time, and 60 dashboard checks parse every one and match the Python checksum to the C one. I re-ran all of it on 27 Oct 2026 and all 874 passed. The board build was 44,484 of 262,144 bytes of flash on 7 Sep.

What I'd do differently

Calibrate the sensors before building on them. The force numbers still come from two placeholder points, so the grams are not real weights. The thermistor constants were never measured off the part either. And with no mat plugged in, D4 floats: it read as about 450 g, an occupied bed and about seven taps a second, and nothing downstream can tell that from a real reading. A pull-down resistor and a fault check for that case would come first next time.

Stack

BoardArduino Nano 33 IoT · SAMD21G18A · Cortex-M0+

FirmwareC · CMSIS registers · FreeRTOS_SAMD21 · ADC · EIC · TC3 · SERCOM0

LaptopPython · Tkinter · SQLite

ChecksC++ host tests · mock registers · simulator · arduino-cli

← back to projects

link SERCOM0 CTRLA = 0x40200106, CTRLB = 0x30000, BAUD = 63019 asked for 115200 baud, got 115219, error 165 ppm against a limit of 20000 ppm D6 to D7 loopback: up pins A0 thermistor, D4 mat, D6 TxD, D7 RxD, D8 red LED, D2 tap scheduler tick 1000 Hz, priorities 0 to 8, heap free 13912 of 14336 mat 5 at 10 ms, command 4 at 50 ms, link 3 at 1 s, temp 1 at 5 min

seconds

force

0 g

319 g

804 g

985 g

1,029 g

1,038 g

1,041 g

121 g

17 g

0 g

10 g

206 g

33 g

520 g

894 g

1,008 g

1,034 g

1,040 g

1,041 g

212 g

208 g

47 g

5 g

0 g

124 g

18 g

0 g

seconds

taps

$AC,SEQ=482,UP=510000,BED=1,T=1934,TB=HYPO,MAT=21,F=0,OCC=0,ALM=02,TAPS=0,ST=27*6E $AC,SEQ=483,UP=511000,BED=1,T=1934,TB=HYPO,MAT=4095,F=120,OCC=0,ALM=02,TAPS=1,ST=27*67 $AC,SEQ=484,UP=512000,BED=1,T=1934,TB=HYPO,MAT=4095,F=357,OCC=0,ALM=02,TAPS=4,ST=27*64 $AC,SEQ=485,UP=513000,BED=1,T=1934,TB=HYPO,MAT=4095,F=405,OCC=0,ALM=06,TAPS=8,ST=27*6C

seconds	taps
0	0
1	0
2	0
3	0
4	0
5	0
6	0
7	0
8	0
9	0
10	0
11	0
12	0
13	3
14	3
15	3
16	3
17	3
18	3
19	3
20	3
21	3
22	3
23	3
24	3
25	3
26	3
27	3
28	3
29	3
30	3
31	3
32	3
33	3
34	3
35	3
36	4
37	5
38	6
39	6
40	6
41	6
42	6
43	6
44	6
45	6
46	6
47	6
48	6
49	6
50	6
51	6
52	6
53	6
54	6
55	6
56	6
57	6
58	6
59	6
60	6
61	7
62	7
63	7
64	7
65	7
66	7
67	10
68	14
69	14
70	14
71	14
72	14
73	14
74	14
75	14
76	14
77	14
78	14
79	14
80	14

seconds	taps
0	0
1	0
2	0
3	0
4	0
5	0
6	0
7	0
8	0
9	0
10	0
11	0
12	0
13	3
14	3
15	3
16	3
17	3
18	3
19	3
20	3
21	3
22	3
23	3
24	3
25	3
26	3
27	3
28	3
29	3
30	3
31	3
32	3
33	3
34	3
35	3
36	4
37	5
38	6
39	6
40	6
41	6
42	6
43	6
44	6
45	6
46	6
47	6
48	6
49	6
50	6
51	6
52	6
53	6
54	6
55	6
56	6
57	6
58	6
59	6
60	6
61	7
62	7
63	7
64	7
65	7
66	7
67	10
68	14
69	14
70	14
71	14
72	14
73	14
74	14
75	14
76	14
77	14
78	14
79	14
80	14

seconds	taps
0	0
1	0
2	0
3	0
4	0
5	0
6	0
7	0
8	0
9	0
10	0
11	0
12	0
13	3
14	3
15	3
16	3
17	3
18	3
19	3
20	3
21	3
22	3
23	3
24	3
25	3
26	3
27	3
28	3
29	3
30	3
31	3
32	3
33	3
34	3
35	3
36	4
37	5
38	6
39	6
40	6
41	6
42	6
43	6
44	6
45	6
46	6
47	6
48	6
49	6
50	6
51	6
52	6
53	6
54	6
55	6
56	6
57	6
58	6
59	6
60	6
61	7
62	7
63	7
64	7
65	7
66	7
67	10
68	14
69	14
70	14
71	14
72	14
73	14
74	14
75	14
76	14
77	14
78	14
79	14
80	14

seconds	taps
0	0
1	0
2	0
3	0
4	0
5	0
6	0
7	0
8	0
9	0
10	0
11	0
12	0
13	3
14	3
15	3
16	3
17	3
18	3
19	3
20	3
21	3
22	3
23	3
24	3
25	3
26	3
27	3
28	3
29	3
30	3
31	3
32	3
33	3
34	3
35	3
36	4
37	5
38	6
39	6
40	6
41	6
42	6
43	6
44	6
45	6
46	6
47	6
48	6
49	6
50	6
51	6
52	6
53	6
54	6
55	6
56	6
57	6
58	6
59	6
60	6
61	7
62	7
63	7
64	7
65	7
66	7
67	10
68	14
69	14
70	14
71	14
72	14
73	14
74	14
75	14
76	14
77	14
78	14
79	14
80	14

seconds	taps
0	0
1	0
2	0
3	0
4	0
5	0
6	0
7	0
8	0
9	0
10	0
11	0
12	0
13	3
14	3
15	3
16	3
17	3
18	3
19	3
20	3
21	3
22	3
23	3
24	3
25	3
26	3
27	3
28	3
29	3
30	3
31	3
32	3
33	3
34	3
35	3
36	4
37	5
38	6
39	6
40	6
41	6
42	6
43	6
44	6
45	6
46	6
47	6
48	6
49	6
50	6
51	6
52	6
53	6
54	6
55	6
56	6
57	6
58	6
59	6
60	6
61	7
62	7
63	7
64	7
65	7
66	7
67	10
68	14
69	14
70	14
71	14
72	14
73	14
74	14
75	14
76	14
77	14
78	14
79	14
80	14

seconds	taps
0	0
1	0
2	0
3	0
4	0
5	0
6	0
7	0
8	0
9	0
10	0
11	0
12	0
13	3
14	3
15	3
16	3
17	3
18	3
19	3
20	3
21	3
22	3
23	3
24	3
25	3
26	3
27	3
28	3
29	3
30	3
31	3
32	3
33	3
34	3
35	3
36	4
37	5
38	6
39	6
40	6
41	6
42	6
43	6
44	6
45	6
46	6
47	6
48	6
49	6
50	6
51	6
52	6
53	6
54	6
55	6
56	6
57	6
58	6
59	6
60	6
61	7
62	7
63	7
64	7
65	7
66	7
67	10
68	14
69	14
70	14
71	14
72	14
73	14
74	14
75	14
76	14
77	14
78	14
79	14
80	14