% LORAPIPE(1) John Goerzen | lorapipe Manual % John Goerzen % October 2019
lorapipe - Transfer data and run a network over LoRa long-range radios
lorapipe [ OPTIONS ] PORT COMMAND [ command_options ]
lorapipe is designed to integrate LoRa long-range radios into a Unix/Linux system. In particular, lorapipe can:
- Bidirectionally pipe data across a LoRa radio system
- Do an RF ping and report signal strength at each end
- Operate an AX.25 network using LoRa, and atop it, TCP/IP
lorapipe is designed to run with a Microchip RN2903/RN2483 as implemented by LoStik.
Drivers for other hardware may be added in the future.
The Microchip firmware must be upgraded to 1.0.5 before running with
lorapipe. Previous versions lacked the radio rxstop
command,
which is a severe limitation when receiving multiple packets rapidly.
See the documents tab for the RN2093 and the firmware upgrade guide - note that the upgrade part is really finicky and you need the "offset" file.
The lorapipe pipe command is the primary one of interest here. It will receive data on stdin, break it up into LoRa-sized packets (see --maxpacketsize), and transmit it across the radio. It also will receive data from the radio channel and send it to stdout. No attempt at encryption or authentication is made; all packets successfully decoded will be sent to stdout. Authentication and filtering is left to other layers of the stack atop lorapipe.
A thin layer atop lorapipe pipe is lorapipe kiss, which implements the AX.25 KISS protocol. It transmits each KISS frame it receives as a LoRa frame, and vice-versa. It performs rudimentary checking to ensure it is receiving valid KISS data, and will not pass anything else to stdout. This support can be used to build a TCP/IP network atop LoRa as will be shown below. Encryption and authentication could be added atop this by using tools such as OpenVPN or SSH.
lorapipe provides only the guarantees that LoRa itself does: that raw LoRa frames which are decoded are intact, but not all frames will be received. It is somewhat akin to UDP in this sense. Protocols such as UUCP, ZModem, or TCP can be layered atop lorapipe to transform this into a "reliable" connection.
It is quite possible to use lorapipe to broadcast data to multiple listeners; an unlimited number of systems can run lorapipe pipe to receive data, and as long as there is nothing on stdin, they will happily decode data received over the air without transmitting anything.
Separate communication channels may be easily achieved by selecting separate radio frequencies.
lorapipe cannot provide collision detection or avoidance, though it does impliement a collision mitigation strategy as described below.
As LoRa radios are half-duplex (they cannot receive while transmitting), this poses challenges for quite a few applications that expect full-duplex communication or something like it. In testing, a particular problem was observed with protocols that use transmission windows and send data in packets. These protocols send ACKs after a successful packet transmission, which frequently collided with the next packet transmitted from the other radio. This caused serious performance degredations, and for some protocols, complete failure.
There is no carrier detect signal from the LoRa radio. Therefore, a turn-based mechanism is implemented; with each frame transmitted, a byte is prepended indicating whether the sender has more data in queue to transmit or not. The sender will continue transmitting until its transmit buffer is empty. When that condition is reached, the other end will begin transmitting whatever is in its queue. This enables protocols such as UUCP "g" and UUCP "i" to work quite well.
A potential complication could arise if the "last" packet from the transmitter never arrives at the receiver; the receiver might therefore never take a turn to transmit. To guard against this possibility, there is a timer, and after receiving no packets for a certain amount of time, the receiver will assume it is acceptable to transmit. This timeout is set by the --eotwait option and defaults to 1000ms (1 second).
The signal about whether or not data remains in the queue takes the form of a single byte prepended to every frame. It is 0x00 if no data will follow immediately, and 0x01 if data exists in the transmitters queue which will be sent immediately. The receiving side processes this byte and strips it off before handing the data to the application. This byte is, however, visible under --debug mode, so you can observe the protocol at this low level.
The Microchip command reference, available at http://ww1.microchip.com/downloads/en/DeviceDoc/40001811A.pdf, describes the parameters available for the radio. A LoRa data rate calculator is available at https://www.rfwireless-world.com/calculators/LoRa-Data-Rate-Calculator.html to give you a rough sense of the speed of different parameters. In general, by sacrificing speed, you can increase range and robustness of the signal. The default initialization uses fairly slow and high-range settings:
sys get ver
mac reset
mac pause
radio get mod
radio get freq
radio get pwr
radio get sf
radio get bw
radio get cr
radio get wdt
radio set pwr 20
radio set sf sf12
radio set bw 125
radio set cr 4/5
radio set wdt 60000
The get
commands will cause the pre-initialization settings to be
output to stderr if --debug
is used. A maximum speed init would
look like this:
sys get ver
mac reset
mac pause
radio get mod
radio get freq
radio get pwr
radio get sf
radio get bw
radio get cr
radio get wdt
radio set pwr 20
radio set sf sf7
radio set bw 500
radio set cr 4/5
radio set wdt 60000
You can craft your own parameters and pass them in with --initfile
to customize the performance of your RF link.
A particular hint: if --debug
shows radio_err
after a radio rx 0
command, the radio is seeing carrier but is getting CRC errors
decoding packets. Increasing the code rate with radio set cr
to a
higher value such as 4/6
or even 4/8
will increase the FEC
redundancy and enable it to decode some of those packets. Increasing
code rate will not help if there is complete silence from the radio
during a transmission; for those situations, try decreasing bandwidth
or increasing the spreading factor. Note that coderate 4/5
to the
radio is the same as 1
to the calculator, while 4/8
is the same as
4
.
Important note: If you have the RN2483-based Lorastik, it requires
a band as part of the mac reset
command. You will need to edit the
config file to say either mac reset 868
or mac reset 433
depending
on which band you will be using. See
#2 for further details.
Although lorapipe pipe doesn't guarantee it preserves application framing, in many cases it does. For applications that have their own framing, it is highly desirable to set their frame size to be less than the --maxpacketsize setting. This will reduce the amount of data that would have to be retransmitted due to lost frames.
As speed decreases, packet size should as well.
The socat(1) program can be particularly helpful; it can gateway TCP ports and various other sorts of things into lorapipe. This is helpful if the lorapipe system is across a network from the system you wish to run an application on. ssh(1) can also be useful for this purpose.
A basic command might be like this:
socat TCP-LISTEN:12345 EXEC:'lorapipe /dev/ttyUSB0 pipe,pty,rawer'
Some systems might require disabling buffering in some situations, or using a pty. In those instances, something like this may be in order:
socat TCP-LISTEN:10104 EXEC:'stdbuf -i0 -o0 -e0 lorapipe /dev/ttyUSB4 pipe,pty,rawer'
For UUCP, I recommend protocol i
with the default window-size
setting. Use as large of a packet size as you can; for slow links,
perhaps 32, up to around 100 for fast, high-quality links. (LoRa seems to
not do well with packets above 100 bytes).
Protocol g
(or G
with a smaller packet size) can also work, but
won't work as well.
Make sure to specify half-duplex true
in /etc/uucp/port
.
Here is an example of settings in sys
:
protocol i
protocol-parameter i packet-size 90
protocol-parameter i timeout 30
chat-timeout 60
Note that UUCP protocol i adds 10 bytes of overhead per packet, so this is designed to work with the default recommended packet size of 100.
Then in /etc/uucp/port
:
half-duplex true
reliable false
ZModem makes a poor fit for LoRa because its smallest block size is 1K. YModem, however, uses a 128-byte block size. Here's an example of how to make it work. Let's say we want to transmit /bin/true over the radio. We could run this:
socat EXEC:'sz --ymodem /bin/true' EXEC:'lorapipe /dev/ttyUSB0 pipe,pty,rawer'
And on the receiving end:
socat EXEC:'rz --ymodem' EXEC:'lorapipe /dev/ttyUSB0 pipe,pty,rawer'
This approach can also be used with many other programs. For
instance, uucico -l
for UUCP logins.
Using the C-kermit distribution (apt-get install ckermit), you can configure for lorapipe like this:
set receive packet-length 90
set send packet-length 90
set duplex half
set window 2
set receive timeout 10
set send timeout 10
Then, on one side, run:
pipe lorapipe /dev/ttyUSB0 pipe
Ctrl-\ c
server
And on the other:
pipe lorapipe /dev/ttyUSB0 pipe
Ctrl-\ c
Now you can do things like rdir
(to see ls from the remote), get
,
put
, etc.
To interact directly with the modem, something like this will work:
cu -h --line /dev/ttyUSB0 -s 57600 -e -o -f --nostop
PPP is the fastest way to run TCP/IP over LoRa with lorapipe. It is subject to a few limitations:
- At most two devices must be using the frequency. PPP cannot support ad-hoc communication to multiple devices like AX.25 can (see below).
- PPP compression should not be turned on. This is because PPP normally assumes a lossless connection, and any dropped packets become rather expensive for PPP to handle, since compression has to be re-set. Better to use compression at the protocol level; for instance, with ssh -C.
To set up PPP, on one device, create /etc/ppp/peers/lora with this content:
hide-password
noauth
debug
nodefaultroute
192.168.2.3:192.168.2.2
mru 1024
passive
115200
nobsdcomp
nodeflate
On the other device, swap the order of those IP addresses.
Now, fire it up on each end with a command like this:
socat EXEC:'pppd nodetach file /etc/ppp/peers/lora,pty,rawer' \
EXEC:'lorapipe --txslot 2000 --initfile=init-fast.txt --maxpacketsize 100 --txwait 120 /dev/ttyUSB0 pipe,pty,rawer'
According to the PPP docs, an MRU of 296 might be suitable for slower links.
This will now permit you to ping across the link. Additional options can be added to add, for instance, a bit of authentication at the start and so forth (though note that LoRa, being RF, means that a session could be hijacked, so don't put a lot of stock in this as a limit; best to add firewall rules, etc.)
Of course, ssh can nicely run over this, and in my testing, PPP was the fastest method of running SSH over LoRa, beating out even AX.25. But then, that makes some sense, since AX.25 has to add addressing bits to every frame since it is a more LAN-like protocol.
The AX.25 protocol was initially designed to be used for amateur radio purposes. As the original amateur radio systems have a number of properties in common with LoRa, it makes a reasonable way to run a TCP/IP stack atop LoRa. lorapipe supports it via the KISS protocol, which is similar to PPP for AX.25.
PPP normally assumes a reliable, point-to-point connection. AX.25 and KISS allow for more than 2 devices to share a frequency.
These instructions assume Debian or Raspbian. Other operating systems may be different.
First, install the AX.25 tools: apt-get install ax25-tools ax25-apps socat
.
Now, edit /etc/ax25/axports
and add a line such as:
lora NODE1 1200 70 1 lorapipe radio
This defines a port named lora, with fake "callsign" NODE1, speed 1200 (which is ignored), maximum packet length 70, and window 1. Keep the packet length less than the --maxpacketsize. It is possible that KISS frames may expand due to escaping; lorapipe will fragment them in this case, but it is best to keep this size significantly less than the lorapipe max packet size to avoid fragmentation as much as possible. On other machines, give them unique callsigns (NODE2 or FOO1 or whatever you like).
Now, start KISS:
kissattach /dev/ptmx lora 192.168.2.2
AX.25 port lora bound to device ax0
Awaiting client connects on
/dev/pts/7
That IP address was made up; you can use any RFC1918 IP address here; just make sure they're different on each node.
It says to connect to /dev/pts/7, so we'll do just that:
socat /dev/pts/7,rawer \
EXEC:'lorapipe /dev/ttyUSB0 kiss,pty,rawer'
Now, assume you connected a second machine to 192.168.2.3, you should be able to ping and talk back and forth between them. Standard commands will work at this stage. You may wish to adjust the packet size in /etc/axports up from 70.
To bring down the link, Ctrl-C the socat sessions and run killall kissattach
.
It should be noted that a TCP ACK encapsulated in AX.25 takes 69 bytes to transmit -- that's a header with no data, and it's 69 bytes! This is a significant overhead. It can be dramatically reduced by using a larger packet size; for instance, in /etc/ax25/axports, thange the packet length of 70 to 1024. This will now cause the --maxpacketsize option to take precedence and fragment the TCP/IP packets for transmission over LoRa; they will, of course, be reassembled on the other end. Setting --txslot 2000 or a similar value will also be helpful in causing TCP ACKs to reach the remote end quicker, hopefully before timeouts expire. --pack may also produce some marginal benefit.
I have been using:
lorapipe --initfile=init-fast.txt --txslot 2000 --pack --debug --maxpacketsize 200 --txwait 150
with success on a very clean (reasonably error-free) link.
For more information, see:
Before lorapipe introduced frame combining and --txslot, performance of SSH over TCP/IP was as low as 25% of its performance over native AX.25. With the addition of the above features, it has achieved parity with native AX.25 on fairly clean links.
There is somewhat more effort on running SSH atop AX.25 natively, since it was not designed to run in such a way. We can make it work, however.
First, on the node which will run the SSH server -- in this example it will be NODE1 -- create an /etc/ax25/ax25d.conf file with contents like this:
NOCALL * * * * * * L
default * * * * * * - root /usr/bin/socat socat -b 220 STDIO TCP:localhost:22
This will cause it to accept connections on AX.25 port 1 (in NODE1-1, the part after the dash is the AX.25 port number), and redirect to local TCP port 22, ssh. The -b 220 assumes the packet length is 220 in /etc/ax25/axports, and causes ssh data to not exceed that length.
Now you can fire it up with ax25d -l.
Connecting to it requires an 8-bit clean AX.25 connection. Unfortunately, axcall(1) does not provide this. ax25_call can, but it must be modified to cause it to not emit the "Connecting..." and "Connected" messages which will confuse ssh. Once done, the connection can be initiated with:
ssh -v -o "ProxyCommand=socat -b 220 STDIO EXEC:'/path/ax25_call -i 220 -o 220 lora NODE2 NODE1-1,pty,rawer'" user@host
NODE2 is the node name that ssh is running on, and NODE1 is the destination node. Replace every instance of 220 here with your maximum packet length.
This is a somewhat fragile setup, and it is recommended to use TCP instead, in general.
lorapipe is a Rust program and can be built by running cargo build --release
. The executable will then be placed in
target/release/lorapipe. Rust can be easily installed from
https://www.rust-lang.org/.
Every invocation of lorapipe requires at least the name of a serial port (for instance, /dev/ttyUSB0) and a subcommand to run.
These options may be specified for any command, and must be given before the port and command on the command line.
-d, --debug : Activate debug mode. Details of program operation will be sent to stderr.
-h, --help : Display brief help on program operation.
--readqual : Attempt to read and log information about the RF quality of incoming packets after each successful packet received. There are some corner cases where this is not possible. The details will be logged with lorapipe's logging facility, and are therefore only visible if --debug is also used.
--pack : Attempt to pack as many bytes into each transmitted frame as possible. Ordinarily, the pipe and kiss commands attempt -- though do not guarantee -- to preserve original framing from the operating system. With --pack, instead the effort is made to absolutely minimize the number of transmitted frames by putting as much data as possible into each.
-V, --version : Display the version number of lorapipe.
--eotwait TIME : The amount of time in milliseconds to wait after receiving a packet that indicates more are coming before giving up on receiving an additional packet and proceeding to transmit. Ideally this would be at least the amount of time it takes to transmit 2 packets. Default: 1000.
--initfile FILE : A file listing commands to send to the radio to initialize it. If not given, a default set will be used.
--txwait TIME : Amount of time in milliseconds to pause before transmitting each packet. Due to processing delays on the receiving end, packets cannot be transmitted immediately back to back. Increase this if you are seeing frequent receive errors for back-to-back packets, which may be indicative of a late listen. Experimentation has shown that a value of 120 is needed for very large packets, and is the default. You may be able to use 50ms or less if you are sending small packets. In my testing, with 100-byte packets, a txwait of 50 was generally sufficient.
--txslot TIME** : The maximum of time in milliseconds for one end of the conversation to continue transmitting without switching to receive mode. This is useful for protocols such as TCP that expect periodic ACKs and get perturbed when they are not delivered in a timely manner. If --txslot is given, then after the given number of milliseconds have elapsed, the next packet transmitted will signal to the other end that it should take a turn. If the transmitter has more data to send, it is sent with a special flag of 2 to request the other end to immediately send back a frame - data if it has some, or a "I don't have anything, continue" frame otherwise. After transmitting flag 2, it will wait up to txwait seconds for the first packet from the other end before continuing to transmit. This setting is not suitable when more than 2 radios are on-frequency. Setting txslot also enables responses to flag 2. The default is 0, which disables the txslot feature and is suitable for uses which do not expect ACKs.
--maxpacketsize BYTES : The maximum frame size, in the range of 10 - 250. The actual frame transmitted over the air will be one byte larger due to lorapipe collision mitigation as described above. Experimentation myself, and reports from others, suggests that LoRa works best when this is 100 or less.
PORT : The name of the serial port to which the radio is attached.
COMMAND : The subcommand which will be executed.
The pipe subcommand is the main workhorse of the application and is described extensively above.
The ping subcommand will transmit a simple line of text every 10 seconds including an increasing counter. It can be displayed at the other end with lorapipe ... pipe or reflected with lorapipe ... pong.
The pong subcommand receives packets and crafts a reply. It is intended to be used with lorapipe ... ping. Its replies include the signal quality SNR and RSSI if available.
John Goerzen [email protected]
I wrote an introduction and a follow-up about TCP/IP on my blog.
Copyright (C) 2019 John Goerzen <[email protected]
This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.