This fork implements interrupt service routine best practice. In the receive interrupt, instead of blocking for whole bytes at a time - voiding any near-realtime behavior of the CPU - only level change and timestamp are recorded. The more time consuming phase detection and byte assembly are done in the main code.
Except at high bitrates, depending on other ongoing activity,
interrupts in particular, this software serial adapter
supports full duplex receive and send. At high bitrates (115200bps)
send bit timing can be improved at the expense of blocking concurrent
full duplex receives, with the SoftwareSerial::enableIntTx(false) function call.
The same functionality is given as the corresponding AVR library but
several instances can be active at the same time. Speed up to 115200 baud
is supported. Besides a constructor compatible to the AVR SoftwareSerial class,
and updated constructor that takes no arguments exists, instead the begin()
function can handle the pin assignments and logic inversion.
It also has optional input buffer capacity arguments for byte buffer and ISR bit buffer.
This way, it is a better drop-in replacement for the hardware serial APIs on the ESP MCUs.
Please note that due to the fact that the ESPs always have other activities ongoing, there will be some inexactness in interrupt timings. This may lead to inevitable, but few, bit errors when having heavy data traffic at high baud rates.
This library supports ESP8266, ESP32, ESP32-S2 and ESP32-C3 devices.
The memory footprint can be optimized to just fit the amount of expected
incoming asynchronous data.
For this, the SoftwareSerial constructor provides two arguments. First, the
octet buffer capacity for assembled received octets can be set. Read calls are
satisfied from this buffer, freeing it in return.
Second, the signal edge detection buffer of 32bit fields can be resized.
One octet may require up to to 10 fields, but fewer may be needed,
depending on the bit pattern. Any read or write calls check this buffer
to assemble received octets, thus promoting completed octets to the octet
buffer, freeing fields in the edge detection buffer.
Look at the swsertest.ino example. There, on reset, ASCII characters ' ' to 'z' are sent. This happens not as a block write, but in a single write call per character. As the example uses a local loopback wire, every outgoing bit is immediately received back. Therefore, any single write call causes up to 10 fields - depending on the exact bit pattern - to be occupied in the signal edge detection buffer. In turn, as explained before, each single write call also causes received bit assembly to be performed, promoting these bits from the signal edge detection buffer to the octet buffer as soon as possible. Explaining by way of contrast, if during a a single write call, perhaps because of using block writing, more than a single octet is received, there will be a need for more than 10 fields in the signal edge detection buffer. The necessary capacity of the octet buffer only depends on the amount of incoming data until the next read call.
For the swsertest.ino example, this results in the following optimized constructor arguments to spend only the minimum RAM on buffers required:
The octet buffer capacity (bufCapacity) is 95 (93 characters net plus two tolerance).
The signal edge detection buffer capacity (isrBufCapacity) is 11, as each
single octet can have up to 11 bits on the wire,
which are immediately received during the write, and each
write call causes the signal edge detection to promote the previously sent and
received bits to the octet buffer.
In a more generalized scenario, calculate the bits (use message size in octets
times 10) that may be asynchronously received to determine the value for
isrBufCapacity in the constructor. Also use the number of received octets
that must be buffered for reading as the value of bufCapacity.
The more frequently your code calls write or read functions, the greater the
chances are that you can reduce the isrBufCapacity footprint without losing data,
and each time you call read to fetch from the octet buffer, you reduce the
need for space there.
The configuration of the data stream is done via a SoftwareSerialConfig
argument to begin(). Word lengths can be set to between 5 and 8 bits, parity
can be N(one), O(dd) or E(ven) and 1 or 2 stop bits can be used. The default is
SWSERIAL_8N1 using 8 bits, no parity and 1 stop bit but any combination can
be used, e.g. SWSERIAL_7E2. If using EVEN or ODD parity, any parity errors
can be detected with the readParity() and parityEven() or parityOdd()
functions respectively. Note that the result of readParity() always applies
to the preceding read() or peek() call, and is undefined if they report
no data or an error.
To allow flexible 9-bit and data/addressing protocols, the additional parity
modes MARK and SPACE are also available. Furthermore, the parity mode can be
individually set in each call to write().
This allows a simple implementation of protocols where the parity bit is used to
distinguish between data and addresses/commands ("9-bit" protocols). First set
up SoftwareSerial with parity mode SPACE, e.g. SWSERIAL_8S1. This will add a
parity bit to every byte sent, setting it to logical zero (SPACE parity).
To detect incoming bytes with the parity bit set (MARK parity), use the
readParity() function. To send a byte with the parity bit set, just add
MARK as the second argument when writing, e.g. write(ch, SWSERIAL_PARITY_MARK).
In general, most pins on the ESP8266 and ESP32 devices can be used by SoftwareSerial, however each device has a number of pins that have special functions or require careful handling to prevent undesirable situations, for example they are connected to the on-board SPI flash memory or they are used to determine boot and programming modes after powerup or brownouts. These pins are not able to be configured by this library.
The exact list for each device can be found in the
ESP32 data sheet
in sections 2.2 (Pin Descriptions) and 2.4 (Strapping pins). There is a discussion
dedicated to the use of GPIO12 in this
note about GPIO12.
Refer to the isValidGPIOpin(), isValidRxGPIOpin() and isValidTxGPIOpin()
functions for the GPIO restrictions enforced by this library by default.
The easiest and safest method is to test the object returned at runtime, to see if it is valid. For example:
#include <SoftwareSerial.h>
#define MYPORT_TX 12
#define MYPORT_RX 13
SoftwareSerial myPort;
[...]
Serial.begin(115200); // Standard hardware serial port
myPort.begin(38400, SWSERIAL_8N1, MYPORT_RX, MYPORT_TX, false);
if (!myPort) { // If the object did not initialize, then its configuration is invalid
Serial.println("Invalid SoftwareSerial pin configuration, check config");
while (1) { // Don't continue with invalid configuration
delay (1000);
}
}
[...]
EspSoftwareSerial is both part of the BSP download for ESP8266 in Arduino,
and it is set up as a Git submodule in the esp8266 source tree,
specifically in .../esp8266/libraries/SoftwareSerial when using a Github
repository clone in your Arduino sketchbook hardware directory.
This supersedes any version of EspSoftwareSerial installed for instance via
the Arduino library manager, it is not required to install EspSoftwareSerial
for the ESP8266 separately at all, but doing so has ill effect.
The responsible maintainer of the esp8266 repository has kindly shared the following command line instructions to use, if one wishes to manually update EspSoftwareSerial to a newer release than pulled in via the ESP8266 Arduino BSP:
To update esp8266/arduino SoftwareSerial submodule to lastest master:
Clean it (optional):
$ rm -rf libraries/SoftwareSerial
$ git submodule update --initNow update it:
$ cd libraries/SoftwareSerial
$ git checkout master
$ git pull