Skip to content

bboettcher3/libmapper

 
 

Repository files navigation

libmapper
libmapper

license:lgpl tag:? downloads:? status:?
forks:? followers:? watchers:?

This library is a system for representing input and output signals on a network and allowing arbitrary "mappings" to be dynamically created between them.

A "mapping", or "connection" associated with relational properties, consists of an Open Sound Control stream being established between source signals and a destination signal – the source is translated to the destination's expected format, with some mathematical expression used to condition the transmitted values as desired. This can be used for example to connect a set of sensors to a synthesizer's input parameters.

To get started quickly with libmapper, be sure to read the tutorials, found in the "doc" folder in this distribution.

History of the mapper project

This project began life as a tool for helping composers and performers to more easily play with custom-built musical instruments, such that a program that reads data from the instrument (e.g. from an Arduino embedded microcontroller) can be connected to the inputs of a sound synthesizer. The first version of this software was written entirely in Cycling 74's Max/MSP, but in order to make it more universally useful and cross-platform, we decided to re-implement the protocol in C; hence, libmapper.

We were already using Open Sound Control for this purpose, but needed a way to dynamically change the mappings – not only to modify the destinations of OSC messages, but also to change scaling, or to introduce derivatives, integration, frequency filtering, smoothing, etc. Eventually this grew into the use of a multicast bus to publish signal metadata and to administrate peer-to-peer connections between machines on a local subnet, with the ability to specify arbitrary mathematical expressions that can be used to perform signal conditioning.

It has been designed with the idea of decentralization in mind. We specifically wanted to avoid keeping a "master" node which is responsible for arbitrating connections or republishing data, and we also wanted to avoid needing to keep a central database of serial numbers or other unique identifiers for devices produced in the lab. Instead, common communication is performed on a multicast UDP port, which all nodes listen on, and this is used to implement a collision-handling protocol that can assign a unique ID to each device that appears. Actual signal data is sent directly from a sender device to the receiver's IP and UDP port.

Advantages of libmapper

This library is, on one level, our response to the widely acknowledged lack of a "query" protocol for Open Sound Control. We are aware for example of [current work on using the ZeroConf protocol for publishing OSC services] (http://sourceforge.net/projects/osctools/), and various efforts to standardize a certain set of OSC address conventions.

However, libmapper covers considerably more ground than simply the ability to publish namespaces, and handles a direct need that we experience in our daily work. We evaluated the use of ZeroConf early on in this work, but we eventually decided that sending OSC messages over multicast was just as easy and required the integration of fewer dependencies. With the addition of being able to control message routing and to dynamically specify signal conditioning equations, we feel this work represents a significant enough effort that it is worth making available to the general public.

It can also be seen as an open source alternative to some commercial products such as Camille Troillard's Osculator and STEIM's junXion, albeit certainly more "barebones" for the moment.

In addition to passing signal data through mapping connections, libmapper provides the ability to query or stream the state of the inputs on a destination device. This permits the use of intermediate devices that use supervised machine learning techniques to "learn" mappings automatically.

A major difference in libmapper's approach to handling devices is the idea that each "driver" can be a separate process, on the same or different machines in the network. By creating a C library with a basic interface and by providing Python and Java bindings, we hope to enable a very wide variety of programming language bindings. Another advantage of a C library is portability: we have demonstrated libmapper working on a Gumstix device, an ethernet-enabled ARM-based microcomputer running Linux that can be easily embedded directly into an instrument.

We also provide "external" Max/MSP and PureData objects, to help integrate libmapper support into programs created in these popular dataflow languages.

Convergent (or many-to-one) mapping refers to situations in which multiple data sources are routed to a single destination. In early versions of libmapper the updates from each source would simple be interleaved, but starting with version 0.4 signal processing can be applied that specifically addresses each source stream and allows arbitrary combining functions.

Known limitations

The "devices and signals" metaphor encapsulated by libmapper is of course not the be-all and end-all of mapping. For signal datatypes we only support homogeneous vectors numbers (integer, float, or double) rather than allowing more complex structured data, since we believe that configurability and complatibility are best served when models are not embedded in the protocol. Using libmapper, complex data structures should be represented as collections of well-labeled signals.

If libmapper is to be used for visual art, transmission of matrixes or even image formats might be interesting but is not currently supported since there is no standard way to represent this in Open Sound Control other than as a binary "blob". It's possible that in this case a protocol such as HTTP that supports MIME type headers is simply a more appropriate choice.

That aside, mapping techniques based on table-lookup or other data-oriented processes don't fit directly into the "connection" paradigm used here. Of course, many of these techniques can easily be implemented as "intermediate" devices that sit between a sender and receiver.

One impact of peer-to-peer messaging is that it may suffer from redundancy. In general it may be more efficient to send all data once to a central node, and then out once more to nodes that request it, at the expense of weighing down the bandwidth of that particular central node. In libmapper's case, if 50 nodes subscribe to a particular signal, it will be repeated that many times by the originating node. Dealing with centralized-vs-decentralized efficiency issues by automatically optimizing decisions on how messages are distributed and where routing takes place is not impossible, but represents non-trivial work--for example, in libmapper the concept of a "router" is actually an independent node that a device sends messages to, which then translates and rebroadcasts; it just happens that the router is embedded in the sender because that was the most efficient place to have it in our scenario. Optimizing of message-passing efficiency/network topology is not a near-term goal, but in the meantime it is entirely possible to use libmapper to explicitly create a centralized network if desired; this will simply imply more overhead in managing connections.

Future plans

We chose to use Open Sound Control for this because of its compatibility with a wide variety of audio and media software. It is a practical way to transmit arbitrary data in a well-defined binary format. In practice, since the actual connections are peer-to-peer, they could technically be implemented using a variety of protocols. High-frequency data for example might be better transmitted using the JACK Audio Connection Kit or something similar. In addition, on embedded devices, it might make sense to "transmit" data between sensors and synthesizers through shared memory, while still allowing the use of the libmapper admin protocol and GUIs to dynamically experiment with mapping.

In the syntax for mathematical expressions, we include a method for indexing previous values to allow basic filter construction. This is done by index, but we also implemented a syntax for accessing previous state according to time in seconds. This feature is not yet useable, but in the future interpolation will be performed to allow referencing of time-accurate values.

The explicit use of timing information may also be useful for certain mapping scenarios, such as "debouncing" signal updates or adding adaptive delays. We are gradually working towards such functionality, developing a syntax for referring to timing data when configuring connection properties and implementing a lighweight synchronization scheme between linked devices that will permit jitter mitigation and correct handling of delays when devices are running on separate computers.

Getting libmapper

The latest version of libmapper source code and binaries can be downloaded from the libmapper website at:

http://libmapper.org

or from the libmapper page on the IDMIL website,

http://idmil.org/software/libmapper

Building and using libmapper

Please see the separate documentation for building libmapper, tutorials on using its API in C, C++, Python, Java, MaxMSP and Pure Data, and doxygen-generated API documentation, in the "doc" directory.

License

This software was written in the Input Devices and Music Interaction Laboratory at McGill University in Montreal, and is copyright those found in the AUTHORS file. It is licensed under the GNU Lesser Public General License version 2.1 or later. Please see COPYING for details.

In accordance with the LGPL, you are allowed to use it in commercial products provided it remains dynamically linked, such that libmapper always remains free to modify. If you'd like to use it in an outstanding context, please contact the AUTHORS to seek an agreement.

Dependencies of libmapper are:

Optional dependencies for the Python bindings:

  • Python, Python license, LGPL-compatible

Used only in the examples:

  • RtAudio, MIT license, LGPL-compatible

About

A library for connecting things.

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • C 76.3%
  • C++ 8.3%
  • C# 5.1%
  • Java 3.4%
  • M4 2.1%
  • Python 1.6%
  • Other 3.2%