A more complete interface for a simple I²C Temperature and Humidity Sensor.
Including:
- access to humidity and temperature as well as status, and info
- access to two 16-bit writable user registers
- full read / write api
- single register and bulk access
- full
Promise/asyncinterface - enable or disable crc (for speed?)
- more complet Modbus error reporting
As always, Adafruit is a good place to start.
The AM2320 support two distinct modes of operation interface. Classic 1-Wire interface, which seems to be the common / preferred method of interaction, and a Modbus interface over I²C.
This library specificly focuses on the I²C interface.
The AM2320 will select the operational mode (1-Wire vs I²C) on power-on. This is achived by forcing the initial state of the Clock (SCL or pin 4) line LOW or HIGH during the power up phase.
The easiest way to achive this is by adding pull-up resistors (to select I²C) to the SCL / SDA lines. While boards like the Raspberry Pi have existing pull-up resistors they may not be sutable for this particular chip.
The pins, when holding gril toward viewer (as above photo), are from left to right: 1, 2, 3, 4 (VDD, SDA, GND, SCL).
📄 static from(bus)
Current usage of rasbus abstract for I²C interface
const { Rasbus } = require('@johntalton/rasbus');
const { Am2320, DEFAULT_ADDRESS } = require('@johntalton/am2320');
const I2C_DEVICE_ID = 1;
Rasbus.i2c.init(I2C_DEVICE_ID , DEFAULT_ADDRESS)
.then(bus => Am2320.from(bus))
.then(device => {
// wake sleep and interact - before time is up
// catch errors liberally within
})
.catch(err => console.log('setup error', err));from takes a second options argument. A value of { check: true }, the default, will enable CRC reads from the bus, and validation on all calls.
📄 wake()
📄 info() (model(), version(), id())
📄 user(), setUser(one, two) (user1(), setUser1(value), user1(), setUser1(value))
📄 bulk() (humidity(), temperature())
TLDR: As noted in the bellow sections the chip is, fickle at times.
To save power and to provide more accurate readings (by not heating up the chip) the AM2320 goes into a deep-sleep. So much so that the I²C interface is put to sleep (which is why many forum post are titled 'not working').
This auto-sleep requires a wake method call prior to interacting with the standard interface. Once woken the chip will responde to I²C commands, however, there is a limited access window to execute command before the chip will return to the sleep state.
While wake is titled as such from the perspective of the api, the bus level call is just a read (specific type, see bellow), and all failures are suppressed (as expected on first wake). The result of this is that "waking" the chip is not a garantee of successfull wake, or that the chip was not already woken (and thus effecting the remaining time for command execution).
While not documented, trying to use a write command more than once per wake period seem to produce failures.
Multiple reads from bulk temperature or humidity will return cached values from captured value at wake.
As example, calling all the bulk access methods for this chip.
return device.wake()
// sleep
.then(() => device.info().then(console.log))
.then(() => device.status().then(console.log))
.then(() => device.user().then(console.log))
.then(() => device.bulk().then(console.log))wake is recommended. Anywhere from 5 to 400 ms has been observed to work. Many standard Promise timeout implementations exist, and a setTimeout wrapper works well.
(note that while some timings work with no wait; others result usefull success ratio, but most fail without a minimum delay. While at the high end, the risk of allowing the chip to return to sleep state incresses).
Reads the model version and 32-bit chip id, independently. Or in a single chip read via info.
Like the above, the status register is (mostly) unused.
Status is a writable byte register, via setStatus. Documented uses are unknown.
Two 16-bit registers user1 and user2 are both read / write. Access in pair is provided.
As an example:
return device.wake()
.then(() => device.temperature().then(({ C }) => device.setUser1(C)))It is of note that these two registers seem to persist accross power cycles.
Temperature and Humidity can be accessed individually or in Bulk (both temperature and humidity in single call).
The values are returned in Celcius / Farenheit (for Temperature) and Percent Relative Humidity (%RH), making interaction easy and flexible
return device.wake()
.then(() => device.bulk())
.then(({ temperature, humidity }) => console.log('results:', temperature.C, '°C', humidity.percent, 'RH%'))Both the read and write register methods are exposed. Any use cases that bypass the above api should be filled as an issue.
The Modbus protocol includes bydirectional CRC 16 (optional disable via from options). This adds a layer of validation to the interactions and a level of confidence in the chips (and this libraries) operation.
Many implemetation bypass the CRC (on both read and write), with some benificial side effects, at the cost of confidence.
While this api provides the promise interface, it suffers from lacking of mechanism to prevent / detect errors from concurrent access. Thus Promise.all() calls to any of the above api calls would likely fail, as each one require multiple underlining I²C write and read operations.
The Modbus api provides for decoupling the implementation such that each payload send / response has some level of handshake. Though this complexity seems overkill and exclusive access to the bus is left for the caller in this case.
The implementation is also tied to the underlining bus system and it behaviors.
Notably if the chip is queried improperly (to fast, not fast enough, for to long, etc) the it may go into a hard fail state. This state requires power cycle to bring the interface back online (no current combination of read/writes seem to effect its state).
Further, the bus layer returns similar error code making error prone to try and evaluate them at this level.
Other limitation of using the /dev/i2c-1 interface (via i2c-bus) present many timing error that could otherwize be mitigated as general exceptions to the bus layer.
As noted in reference to wake, the timing is left much to the caller. However, the noted nature of the transaction make this timing prone to external factors, and thus a wider range of enviroment testing is needed.
A note should be made of the delicate intermix of I²C write readBuffer (in read) and writeBuffer readBuffer (in write) as well as the use of read in the wake command. The use of the SMBus command set, and the lack of consistency indicate poor understanding of chips operation.
While high level api calls (such as info, bulk, humidity etc) are the intended public interface, the Modbus read / write are also exposed.
And while the high level api should not produce these errors, they can still be exposed when parsing unexpected / invalid bus data as a result of malformed or corrupted interaction.
Some standard rules form the datasheet:
- Reading more then allowed (ten registers / bytes)
return device.wake()
.then(() => device.read(0x00, 11))
Error: ModBus error message: ILLEGAL_ADDRESS- Write to the
statusregister (a single byte register), and attempt to write touser116-bit register, within a single call. While both (all three) are writable registers, thestatusregister is explicitly to be wrien alone (Further enforcing the usfullness of thesetStatusmethod).
return device.wake()
.then(() => device.write(0x0F, [0b00000001, 0x00, 0x25]))
Error: ModBus error message: WRITE_DISABLED- Another WRITE_DISABLED failure comes when writing to address that is not writable. In this case we start at a writable address, but overflow (a correct write to the block would have used
USER_1_HIGH).
return device.wake()
.then(() => device.write(REGISTERS.USER_1_LOW, [0x00, 0x2A, 0x00, 0x25]))
Error: ModBus error message: WRITE_DISABLED- The last address is
0x1F(Retention 😄). Read from the vally beyond
return device.wake()
.then(() => device.read(0x2A, 1))
Error: ModBus error message: WRITE_DATA_SCOPE- Write more then allowed (ten registers / bytes)
return device.wake()
.then(() => device.write(REGISTERS.USER_1_HIGH, [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10]))
Error: ModBus error message: WRITE_DATA_SCOPEThe bus / chip failures seem to be common. while some notes about single write per wake sesion and limiting number of reads and the timing between the have been noted above.
Other failure cases seems to exist
- the
wake/bulkpoll cycle seems to fall into a toggling success / failure state - long runs of calls (1K calls to
bulkin a row) can knock the chip offline, requiring power cycle - all timing error represnet themselves as critical bus failures