|Ethernet data rates||Fast Ethernet|
|Minimum latency||1 1⁄3 ms|
|Maximum channels per link||64|
|Maximum sampling rate||96 kHz|
|Maximum bit depth||24 bits|
CobraNet is a combination of software, hardware, and network protocols designed to deliver uncompressed, multi-channel, low-latency digital audio over a standard Ethernet network. Developed in the 1990s, CobraNet is widely regarded as the first commercially successful implementation of audio over Ethernet.
CobraNet was designed for and is primarily used in large commercial audio installations such as convention centers, stadiums, airports, theme parks, and concert halls. It is most useful in applications where a large number of audio channels must be transmitted over long distances or to multiple locations.
CobraNet is an alternative to analog audio, which suffers from signal degradation over long cable runs due to electromagnetic interference, high-frequency attenuation, and voltage drop. Additionally, the use of digital multiplexing allows audio to be transmitted using much less cabling than analog audio.
CobraNet was developed in 1996 by Boulder, Colorado-based Peak Audio. Initial demonstrations were of a 10 Mbit/s point-to-point system with limited channel capacity. The first permanent installation of CobraNet in this early form was to provide background music throughout the Animal Kingdom theme park. The first commercial use of CobraNet as an interoperable standard was during the half-time show at Super Bowl XXXI in 1997.
CobraNet was first introduced as an interoperable standard in collaboration with manufacturer QSC Audio Products. QSC was the first to license the technology from Peak Audio and marketed it under the RAVE brand. At this point CobraNet had graduated to fast Ethernet and used a unique collision avoidance technique to carry up to 64 channels per Ethernet collision domain.
CobraNet was subsequently enhanced to support and eventually require a switched Ethernet network. An SNMP agent was added for remote control and monitoring. Support for higher sample rates, increased bit resolutions and lowered latency capabilities were later introduced in an incremental and backwards-compatible manner.
In May 2001, Cirrus Logic announced that it had acquired the assets of Peak Audio. Leveraging Cirrus DSP technology, a low-cost SoC implementation of CobraNet was developed and marketed. CobraNet has been widely adopted by commercial audio equipment manufacturers and is used in many facilities.
Advantages and disadvantages
- Cabling cost – using CobraNet and fast Ethernet, 64 channels of uncompressed digital audio are carried through a single, inexpensive category 5 cable. In the analog world, this would have required 64 separate analog audio cables, each of which cost the same or more than the cat-5 cable. Using gigabit and/or fiber optic Ethernet variants, the cost of cabling per audio channel is further reduced compared to the fast Ethernet connections. Also, since CobraNet data can coexist with data traffic over existing Ethernet networks, a single network infrastructure can serve audio distribution and other networking needs.
- Flexibility – the network provides flexibility for future changes to the system. For instance, audio routing can be changed on the fly with network commands, and do not require any rewiring.
- Reliability – use of Ethernet by CobraNet offers many high availability features such as Spanning Tree Protocol, link aggregation, and network management. For critical applications, CobraNet devices can be wired with redundant connections to the network. In this configuration, if one CobraNet device, cable, or Ethernet switch fails, the other takes over almost immediately.
- Audio quality – audio is transmitted in digital form, and therefore provides reduced susceptibility to electromagnetic interference, crosstalk, coloration, and attenuation owing to cable impedance.
- Latency – delays over the CobraNet transmission medium itself are at least 1 1⁄3 milliseconds[note 1] per network traversal. For some applications, these delays can be unacceptable especially when combined with further delays resulting from propagation time, digital signal processing and the conversions between analog and digital.
- Hardware cost – although significant money is usually saved in cabling, at least part of that money is spent on the required CobraNet interfaces which encode and decode the CobraNet signal.
CobraNet is transmitted using standard Ethernet packets. Instead of using TCP/IP packets, CobraNet transfers data using data link layer packets, which travel quickly through hubs, bridges and switches, and are not as susceptible to the latency and QoS problems commonly found in streaming protocols using a higher transport layer. However, since CobraNet does not use IP protocol, its packets cannot travel through routers, and therefore it is limited to use on a LAN; CobraNet cannot be used over the Internet. The network over which CobraNet is transmitted must be able to operate at a minimum of 100 Mbit/s. All CobraNet packets are identified with a unique Ethernet protocol identifier (0x8819) assigned to Cirrus Logic.
CobraNet is not designed to work over wireless networks. Bandwidth and reliability issues associated with typical 802.11 wireless networks tend to cause frequent dropouts and errors. However, wireless communication of CobraNet data can be reliably accomplished using lasers.
Channels and bundles
CobraNet data is organized into channels and bundles. A typical CobraNet signal can contain up to 4 bundles of audio travelling in each direction, for a total of 8 bundles per device. Each bundle houses up to 8 channels of 48 kHz, 20-bit audio, for a total capacity of 64 channels. CobraNet is somewhat scalable, in that channel capacity increases when 16-bit audio is used, and channel capacity decreases when 24-bit audio is used. The number of channels allowed per bundle is limited by the 1,500-byte Ethernet MTU.
There are three types of bundles: multicast, unicast, and private:
- Multicast bundles are sent from one CobraNet device to all other CobraNet devices in the network using Ethernet multicast addressing. Each CobraNet device individually determines if it will use the bundle or discard it. Therefore, multicast bundles are more bandwidth-intensive than other bundle types. Bundle numbers 1–255 are reserved for multicast bundles.
- Unicast bundles are sent from one CobraNet device to any other device or devices configured to receive the bundle number. Unicast bundles are much more efficient because network switches route them only to devices which actually want to receive them. Despite their name, unicast bundles may still be sent to multiple devices, either by transmitting multiple copies of the audio data or using multicast addressing. Bundle numbers 256–65279 are reserved for unicast bundles.
- Private bundles may be sent with unicast or multicast addressing. Bundle numbers 65280–65535 are reserved for private bundles. Private bundle numbers are paired with the MAC address of the device that transmits them. To receive a private bundle, both the bundle number and the MAC address of the transmitter must be specified. Because 256 private bundles available to each transmitter, there is no limit on the total number private bundles on a network.
As long as multicast bundles are used sparingly, it is virtually impossible to exceed the bandwidth of a 100 Mbit network with CobraNet data. However, there are limitations to the maximum number of bundles that can be sent on a network, since the conductor must include data in its beat packets for every bundle on the network, and the beat packet is limited to 1,500 bytes. If each device is transmitting one bundle, there may be up to 184 transmitters active simultaneously (for a total of 184 bundles). If each device is transmitting four bundles, then only 105 transmitters can be active, although they would be producing a total of 421 active bundles. The use of private bundles does not require any additional data in the beat packet, so these network limitations can be sidestepped by using private bundles.
The CobraNet network is synchronized to a single CobraNet device known as the conductor. A conductor priority can be configured to influence selection of the conductor. Among devices with the same conductor priority, the first to establish itself on the network becomes is elected conductor. All other devices are known as performers. In the event that the conductor fails, another CobraNet device will be chosen to become the conductor within milliseconds. CobraNet cannot function without a conductor.
Four main types of packet are used in the transmission and synchronization of CobraNet:
- Beat packets – the conductor outputs a beat packet to all other CobraNet devices on the network at a rate of 750 packets per second. All other CobraNet devices on the network synchronize their audio clock and their data transmissions to the beat packet. The beat packet contains network operating parameters, clock data and transmission permissions for multicast and unicast bundles.
- Audio packets – also known as isochronous data packets, these packets are sent out by all CobraNet devices after they receive a beat packet. At standard latency settings, one audio packet is sent for each beat packet received, and each audio packet includes 64 samples of audio data per channel. At lower latency settings, audio packets may be sent twice or four times for each beat packet received. Bundles do not share packets; separate packets are sent in sequence for each bundle transmitted from the same device.
- Reservation packets – these packets are transmitted as needed or typically once per second at minimum. Their function is to control bandwidth allocation, initiate connections between CobraNet devices, and monitor the status of CobraNet devices.
- Serial bridge packets – asynchronous serial data may be sent between CobraNet devices on the same network. Many standard asynchronous serial formats are supported, including RS-232, RS-422, RS-485 and MIDI.
The buffering and transmission of audio data in Ethernet packets typically incurs a delay of 256 samples or 5 1⁄3 milliseconds. Additional delays are introduced through A-D and D-A conversion. Latency can be reduced by sending smaller packets more often. In most cases, the programmer can choose the desired CobraNet latency for a particular CobraNet device (5 1⁄3, 2 2⁄3, or 1 1⁄3 milliseconds). However, reducing audio latency has consequences:
- Reducing latency requires more processing by the CobraNet interface and may reduce channel capacity.
- Reducing latency places additional demands on network performance, and may not be possible in some network configurations if the forwarding delay is too great.
- Since reducing latency means sending smaller packets more often, more high resolution (i.e. 96 kHz, 24-bit) audio channels can be sent per bundle without exceeding the 1,500-byte payload limit for Ethernet packets. See the table below for bundle capacity limits:
|Latency||Channels per bundle|
|16-bit, 48 kHz||20-bit, 48 kHz||24-bit, 48 kHz||16-bit, 96 kHz||20-bit, 96 kHz||24-bit, 96 kHz|
|5 1⁄3 ms||8||8||7||5||4||3|
|2 2⁄3 ms||8||8||8||8||8||7|
|1 1⁄3 ms||8||8||8||8||8||8|
It may seem from the table above that more information can be sent at a lower latency. However, that is not the case. More channels can be sent per bundle, but fewer bundles can be processed simultaneously by one device. So, while eight 24-bit, 96 kHz channels can be sent in one bundle at 1 1⁄3 ms latency, due to processing constraints, the CobraNet device may only be able to send and receive one bundle instead of the usual four. The bundle capacity of CobraNet devices are unique to the particular device, and are not always the same. However, below is a table illustrating the bundle capacity for a Biamp AudiaFLEX-CM DSP device. The Rx and Tx columns indicate the absolute maximum number of channels that can be received or transmitted. The Rx/Tx column represents the maximum number of channels that can be received and transmitted simultaneously.
|Channels per bundle||5 1⁄3 ms latency||2 2⁄3 ms latency||1 1⁄3 ms latency|
Hardware and software
CobraNet network cards
CobraNet interfaces come in several varieties, some of which can support more channels than others. Additionally, CobraNet interfaces have two Ethernet ports labelled "primary" and "secondary". Only the primary Ethernet port needs to be connected, but if both ports are connected they become a redundant failsafe. That is, if the primary port loses communication, the secondary port quickly takes over. Careful network design and topology which takes advantage of this feature can provide extremely high reliability in critical applications.
The typical CobraNet interfaces provided by Cirrus Logic are the CM-1 and the CM-2:
- CM-1 – the standard CobraNet card, provides 32 in and 32 out audio channels.
- CM-2 – compact, low-power, lower cost design provides 8 or 16 audio channels.
Both cards are designed to be added to audio products by the manufacturer.
Cirrus Logic provides a software application known as CobraCAD, which assists in the design of the network on which the CobraNet system will run. It helps to identify if there are too many routers between two CobraNet devices, if a certain latency is possible given the network configuration, and other tasks. However, Cirrus Logic does not provide software to manipulate their hardware. In fact, in the simplest of cases, no software is required by the end user. For instance, a simple breakout box which converts a CobraNet signal to eight analog audio signals would require little or no configuration by the end user apart from possibly selecting the bundle number. If configuration is required (for example, in a DSP box with integrated CobraNet I/O), then the manufacturer of the device typically supplies proprietary software for that purpose.
- CobraNet supports three latency modes: 1 1⁄3, 2 2⁄3 and 5 1⁄3 ms. See #Latency for details.
- "Best Practices in Network Audio". Audio Engineering Society. 2009. Retrieved 2010-05-05.
- Karagosian, Michael (2004), Following the Digital Audio Chain, retrieved 2007-03-19
- The back of the net, ProAudio-Central, 2 August 2010, retrieved 2010-08-17
- Karagosian, Michael (2006), How Theme Parks Work (Part 3:Networks), retrieved 2007-03-19
- Audio Networking (2009), AARC-NET, Audio Networking Made Simple, archived from the original on 2011-07-11
- US patent 5761430, "Media access control for isochronous data packets in carrier sensing multiple access systems"
- Doering, Christian (2001), Fiber in the Whole (House): Cirrus Logic Buys Peak Audio, retrieved 2009-11-30
- Cirrus Logic, Ethernet Overview, retrieved 2009-12-01
- Cirrus Logic, CobraNet FAQ, Question 12, retrieved 2009-12-01
- Cirrus Logic, Inc. (2–2006). CobraNet Programmer's Reference (PDF). 2.5. pp. 7–27. Retrieved 2009-11-30.
- Yamaha System Solutions (2006). An introduction to networked audio (PDF). p. 7. Retrieved 2009-12-01.
- Yamaha System Solutions (2006). Networked audio system design with CobraNet (PDF). p. 4. Retrieved 2009-12-01.
- Gross, Kevin. Digital Audio Distribution Systems. Retrieved 2009-12-01.
- Renkus Heinz, Inc. Renkus-Heinz ST Series (PDF). p. 4. Retrieved 2009-12-01.
- Cirrus Logic, CobraNet FAQ, Question 13, retrieved 2009-11-30
- "Whirlwind E-Beam Laser". Retrieved 2010-09-18.
- Cirrus Logic, CobraNet FAQ, Question 28, retrieved 2009-11-30
- Cirrus Logic, CobraNet FAQ, Question 24, retrieved 2009-11-30
- Biamp Systems (2007-02-14). Audia Operation Manual (PDF). p. 128. Retrieved 2009-11-30.
- Cirrus Logic, CobraNet Networked Digital Audio, retrieved 2007-03-19
- Whirlwind. CI8M user manual (PDF). p. 1. Retrieved 2010-09-18.