This paper describes the design and evolution of three multi-speaker diffusion systems that provide 3d surround capabilities, and are used for research, composition and concert diffusion at CCRMA, Stanford University.== The Listening Room was built when our building was completely remodeled in 2005 and was equipped with a full 3d surround system, a 16 speaker array in a 4+8+4 configuration driven from a digital mixer. It became clear that a digital mixer was not the best interface for this multi-channel system, so we implemented a custom system based on our Linux workstations.The initial design was based on a workstation controlled through a couple of USB fader boxes. We wrote the software in SuperCollider, and also used additional free software packages such as Jconvolver and Ambdec. The system was designed to boot unassisted into the control software.Audio interfacing was provided by two PCI RME audio interfaces with MADI and ADAT I/O and D/A A/D converters. Users could connect through several interfaces, including ethernet.The system proved to be stable, and it was expanded in 2011 to a 22.4 system. The selected spatial configuration was 1+6+8+6+1, which could provide decoding of 3rd order Ambisonicss.== Through CCRMA's history immersive music production and subsequent concerts have been part of daily life. Concerts were staged in the D.C. Power Lab and later the Frost auditorium using a four channel system, also in Dinkelspiel and Campbel Recital Hall and our own Ballroom, and later in our backyard at The Knoll using arrays of up to 8 speakers driven from analog and later digital mixers.The planned construction of a new concert hall at Stanford spurred us to start planning events for the inaugural season (2013). Starting in 2009 we gradually started buying new speakers for our concert system. Outdoor concerts in 2011 and 2012 were testing grounds for new diffusion technologies and in 2012 we stopped using a digital mixer as the control system and successfully transitioned to one of our Linux workstations as the diffusion engine, using all free software components and custom programming.In 2013 we staged several events in the new Bing Main Hall and its rehearsal space, the Studio. In both cases with dome speaker configurations of 12+8+4+1 which enabled us to render pieces with the same flexibility we enjoyed in our Listening Room, but for a big audience.The system, later called the GRAIL (the Giant Radial Array for Immersive Listening) evolved over the years and can be deployed with up to 32.8 speakers.= The Stage, our small in-house concert hall, also built during the 2005 renovation, also evolved over the years, starting with an 8 channel ring of mid-field monitors that used to be in our Ballroom space. The system was upgraded in 2009 to 16 channels in a two ring configuration, with the upper speakers later repositioned for better rendering of 3d soundfields.Over the past two years the system has been completely remodeled, and while the original 16 channel system was kept as an operating mode of the full system, the array has been expanded to 56 speakers and 8 subwoofers, which can render up to 6th order Ambisonics. Again, a Linux-based control computer controls diffusion, like in the Listening Room.The selection of an audio technology that could support this expansion was crucial, and AVB (Audio Video Bridging) was the only viable alternative we found that could work with fully free and open systems. The audio system consists of 8 audio interfaces (Motu AVB devices), interconnected through AVB. 64 channel connection points to control and user computers are supported through USB.== AVB also allowed us to expand our GRAIL system beyond our original 32 channel maximum. This not only makes the system configurable and extensible, but also facilitates cabling as interfaces can be deployed where they are needed to minimize cabling.== Another upgrade of the Listening Room happened very recently, and upgraded all its speakers (now in a 22.8 configuration). We also replaced the control computer and audio interfaces with new hardware and software, this time based again on Motu and AVB hardware. The system is now easier to use as its main interface is just USB interfaces instead of a now cumbersome ethernet based Netjack system.== ConclusionThe design and evolution of all three systems will be described in more detail in the finished article. All three systems are in constant use by students, composers and researchers.