This tutorial presents the GNU Radio ZMQ blocks. It is a set of six Source Blocks and six Sink Blocks. The naming convention follows other source and sink blocks in that a source block provides data entering a GNU Radio flowgraph and a sink block sends data out of the flowgraph. It is a flowgraph-oriented perspective.
From the
ZeroMQ
website: "ZeroMQ (also known as ØMQ, 0MQ, or zmq) looks like an embeddable networking library but acts like a concurrency framework. It gives you sockets that carry atomic messages across various transports like in-process, inter-process, TCP, and multicast."
For GNU Radio, the two basic groups of ZMQ blocks are those which transport stream data, and those which transport messages. They are described below.
ZMQ blocks come in pairs:
PUB - SUB
PUSH - PULL
REQ - REP
The PUB, PUSH, and REP blocks are always sink blocks; the others are source blocks. Choosing which pair to use depends on your system architecture.
The PUB - SUB pair can be compared to broadcasting. The PUBlish sink sends out data which can be received by one or more SUBscribers.
The PUSH - PULL pair is a point to point link of equal peers.
The REQ - REP pair is a point to point link which operates in lock-step: one REQuest input gives one REPly output. This case changes the perspective somewhat in that the flowgraph is acting as a server for a remote client.
Data Blocks
ZMQ data blocks transport raw stream data; there is no formatting. The data type and the sample rate are determined by the flowgraph feeding the ZMQ Sink. Therefore the flowgraph or program receiving the data must know those parameters in order to interpret the data correctly.
Users of ZMQ blocks are expected to have some familiarity with ZeroMQ. In particular, one should be cognizant of the differences between ZMQ sockets and BSD sockets. See the
ZMQ Socket API
for an overview.
ZMQ blocks use
endpoints
to describe how ZMQ should pass the data. While the most common endpoint uses TCP to transfer the data, other protocols are possible. See
zmq_tcp
and
zmq_ipc
for a description of how each protocol defines endpoints.
The cases described below explain the differences in the block addressing. To conform to port addressing defined by the
Internet Assigned Numbers Authority (IANA)
, private ports can be assigned in the range 49152–65535. Within a single flowgraph using ZMQ blocks is not recommended, since
Virtual_Source
and
Virtual_Sink
blocks are much more efficient.
TCP Bind vs Connect
Some users might be tempted to connect directly to GNU Radio ZMQ Blocks. While this is possible, some caution is needed. First be aware that in any topology, there must be
exactly one
bind
to a given endpoint, while there may be multiple
connect
s to the same endpoint. In GNU Radio, stream
sinks
bind
and stream
sources
connect
. Message blocks accept a parameter that specify whether the block should
bind
or
connect
.
Also be aware that the semantics of TCP endpoints vary between
bind
and
connect
.
TCP Bind
When binding a TCP endpoint, you specify where you will listen for connections. If you specify an IP address, you are telling the socket to only accept connections on the network associated with that address (e.g.,
127.0.0.1
or
192.168.1.123
). In some cases, you may want to listen on all networks connected to your node. For GNU Radio, you should use
0.0.0.0
as the wildcard address; although ZMQ does accept
*
as a wildcard, it doesn't work well in all cases. So, you may choose to
bind
to
tcp://0.0.0.0:54321
.
Be aware that if you don't enter an IP address,
bind
treats the value as a network adapter name (e.g.,
eth0
). See
zmq_tcp
.
Bind to
tcp://localhost:1234
will not do what you think!
Use either
tcp://127.0.0.1:1234
or
tcp://lo:1234
(for the loopback network adapter).
TCP Connect
When connecting a TCP endpoint, you specify the remote endpoint you want to connect to. You can specify either an IP address or a DNS resolvable name. This difference in semantics between
connect
and
bind
is confusing, but must be honored, in order to have your flowgraph communicate as you expect. The simplest solution is to use IP addresses everywhere, but that is not an option in some configurations (e.g., deploying into a Kubernetes cluster).
Wire Format
The ZMQ stream blocks have the option to pass tags. In addition, the PUB/SUB blocks support filtering. Both of these options affect the ZMQ wire protocol.
When a filter string is supplied to a PUB/SUB block, GNU Radio uses
multi-part messages
to send the filter string, followed by the payload. Non-GNU Radio code attempting to interface with GNU Radio ZMQ blocks must be prepared for this part, and discard it. Note that the sender only sends this message part if a non-empty filter has been specified.
Next, if sending tags is enabled, any tags within the window of the data to be sent are encoded in a special format and
prepended
to the payload data. If tags are not enabled, this header is elided.
These two features make matching the sender configuration to the receiver configuration essential. Failure to do so will cause runtime errors in your flowgraph.
Examples
Separate GR flowgraphs on Same Computer
When the ZMQ blocks are in separate flowgraphs but on the same computer, the IP address should be
127.0.0.1
for localhost. It has less overhead than a full IP.
If the Source and Sink blocks are on two different computers, then the IP and port number of the Sink block must be specified on each end of that connection. For example, if the Sink is on IP
192.168.1.194:50241
and the Source is on IP
192.168.1.85
, both Source and Sink blocks must specify the Sink IP and port
192.168.1.194:50241
.
Python Program as a REQ / REP Server
The following Python program receives a string message on its
REQ
uest socket, capitalizes the text, and sends the string on its
REP
ly socket. The terminology gets confusing here because the incoming REQ came from a GR
ZMQ_REP_Message_Sink
and is returned to a
ZMQ_REQ_Message_Source
. Just remember that a
sink
is the terminating point of a flowgraph (swallows all data) and a
source
is the origin of a flowgraph (produces or ingests data).
#!/usr/bin/python3
# -*- coding: utf-8 -*-
# zmq_REQ_REP_server.py
# This server program capitalizes received strings and returns them.
# NOTES:
# 1) To comply with the GNU Radio view, messages are received on the REQ socket and sent on the REP socket.
# 2) The REQ and REP messages must be on separate port numbers.
import pmt
import zmq
_debug = 0 # set to zero to turn off diagnostics
# create a REQ socket
_PROTOCOL = "tcp://"
_SERVER = "127.0.0.1" # localhost
_REQ_PORT = ":50246"
_REQ_ADDR = _PROTOCOL + _SERVER + _REQ_PORT
if (_debug):
print ("'zmq_REQ_REP_server' version 20056.1 connecting to:", _REQ_ADDR)
req_context = zmq.Context()
if (_debug):
assert (req_context)
req_sock = req_context.socket (zmq.REQ)
if (_debug):
assert (req_sock)
rc = req_sock.connect (_REQ_ADDR)
if (_debug):
assert (rc == None)
# create a REP socket
_PROTOCOL = "tcp://"
_SERVER = "127.0.0.1" # localhost
_REP_PORT = ":50247"
_REP_ADDR = _PROTOCOL + _SERVER + _REP_PORT
if (_debug):
print ("'zmq_REQ_REP_server' version 20056.1 binding to:", _REP_ADDR)
rep_context = zmq.Context()
if (_debug):
assert (rep_context)
rep_sock = rep_context.socket (zmq.REP)
if (_debug):
assert (rep_sock)
rc = rep_sock.bind (_REP_ADDR)
if (_debug):
assert (rc == None)
while True:
# Wait for next request from client
data = req_sock.recv()
message = pmt.to_python(pmt.deserialize_str(data))
print("Received request: %s" % message)
output = message.upper()
# Send reply back to client
rep_sock.send (pmt.serialize_str(pmt.to_pmt(output)))
Python Program as a PUSH / PULL Server
Similar to the example above, the following Python program receives a string message on its ZMQ PULL socket, capitalizes the text, and returns the string on its ZMQ PUSH socket.
#!/usr/bin/python3
# -*- coding: utf-8 -*-
# zmq_PUSH_PULL_server.py
import sys
import pmt
import zmq
_debug = 0 # set to zero to turn off diagnostics
# create a PUSH socket
_PROTOCOL = "tcp://"
_SERVER = "127.0.0.1" # localhost
_PUSH_PORT = ":50252"
_PUSH_ADDR = _PROTOCOL + _SERVER + _PUSH_PORT
if (_debug):
print ("'zmq_PUSH_PULL_server' version 20068.1 binding to:", _PUSH_ADDR)
push_context = zmq.Context()
if (_debug):
assert (push_context)
push_sock = push_context.socket (zmq.PUSH)
if (_debug):
assert (push_sock)
rc = push_sock.bind (_PUSH_ADDR)
if (_debug):
assert (rc == None)
# create a PULL socket
_PROTOCOL = "tcp://"
_SERVER = "127.0.0.1" # localhost
_PULL_PORT = ":50251"
_PULL_ADDR = _PROTOCOL + _SERVER + _PULL_PORT
if (_debug):
print ("'zmq_PUSH_PULL_server' connecting to:", _PULL_ADDR)
pull_context = zmq.Context()
if (_debug):
assert (pull_context)
pull_sock = pull_context.socket (zmq.PULL)
if (_debug):
assert (pull_sock)
rc = pull_sock.connect (_PULL_ADDR)
if (_debug):
assert (rc == None)
while True:
# Wait for next request from client
data = pull_sock.recv()
message = pmt.to_python(pmt.deserialize_str(data))
# print("Received request: %s" % message)
output = message.upper() # capitalize message
# Send reply back to client
push_sock.send (pmt.serialize_str(pmt.to_pmt(output)))
Python Program to Process Flowgraph Data
Here's the code to do GNU Radio --> Python over ZMQ PUB/SUB, which is by far the most useful case. Often you use GNU Radio for the signal processing but then at some point you want the resulting stream to go to a regular Python app. PUB/SUB is just so you can have multiple apps getting the stream. In the ZMQ_PUB_Sink you can switch * to 127.0.0.1, but with *, any device on the LAN would be able to see it. It's essentially broadcasting to any interface, not just the loopback.
An example flowgraph:
and the Python code:
#!/usr/bin/python3
# -*- coding: utf-8 -*-
# zmq_SUB_proc.py
# Author: Marc Lichtman
import zmq
import numpy as np
import time
import matplotlib.pyplot as plt
context = zmq.Context()
socket = context.socket(zmq.SUB)
socket.connect("tcp://127.0.0.1:55555") # connect, not bind, the PUB will bind, only 1 can bind
socket.setsockopt(zmq.SUBSCRIBE, b'') # subscribe to topic of all (needed or else it won't work)
while True:
if socket.poll(10) != 0: # check if there is a message on the socket
msg = socket.recv() # grab the message
print(len(msg)) # size of msg
data = np.frombuffer(msg, dtype=np.complex64, count=-1) # make sure to use correct data type (complex64 or float32); '-1' means read all data in the buffer
print(data[0:10])
# plt.plot(np.real(data))
# plt.plot(np.imag(data))
# plt.show()
else:
time.sleep(0.1) # wait 100ms and try again
