Chip developer Marvell has announced a series of products that it hopes will unlock the market for key elements within the vRAN and O-RAN architecture.
The company has announced a complete Radio Unit (RU) platform, in partnership with Analog Devices and Benetel, as well as L1 software and O-RAN features for a vDU, and support for a CU platform.
Raj Singh, Executive Vice President Processors Business Group at Marvell Semiconductor, told TMN:
“What we’re doing is taking all of our years of experience and expertise and saying, ‘Where are the choke points in O-RAN. What is stopping this market from taking off?” And two things come to light.
“One is that RadioHeads in particular in the Open RAN market are 2T2R, 4T4R – there’s really no massive MIMO (mMIMO) RadioHeads available today. There are some various initiatives – Mavenir has made some announcements, Fujitsu is trying to do something with Rakuten – all of which are based on FPGAs with the attendant issues on what you need to cool it, what the price points are, and what power you are going to burn from an opex viewpoint.
“And then the other big one is the lack of L1 software in particular that is hampering the market. So what we are announcing is in three segments. First is a massive MIMO RU reference design.
“What we are announcing here is complete manufacturable design in conjunction with Analog Device and Benetel in Ireland, which does 32T/32R massive MIMO, 8 layer, 100MHz channelisation. It has software for the lower L1 and beam forming, including for O-RAN management and control plane, and then the new generation of ADI transceivers have built-in DFE. So this is the entire thing – a complete system including the antenna array, the chassis and the metalwork, and then customers can go to ODMs to manufacture these. We have several ODMs lined up to manufacture these. So this is a big step in jump-starting the radio part of the O-RAN ecosystem – no-one’s building [O-RAN] mMIMO radioheads today because it’s a really hard thing to do.”
Benetel’s CE Adrian O’Connor said to TMN that the company expected the mMIMO RU to be available in the 2nd Half of 2021. He said, “The engagement with ADI and Marvell around mMIMO builds on a long standing relationship with both companies. Benetel is bringing the system solution experience to provide a reference product and plans to sell the resultant product to the market.”
The RU and new DU designs are powered by Marvell’s new platform, Fusion O, in combination with new reference software that includes O-RAN compliant beam forming, fronthaul, timing and compression specifications.
As well as the mMIMO RU, Singh said: “We are also providing silicon for a DU – both an O-RAN and standard DU, and in addition to that we are providing L1 reference software. So that is ready to start integrating to L2. And then we have the CU deploying our Octeon TX2 processor along with several libraries to do full virtualisation.”
Singh said that current approaches to vRAN that rely on software-only approaches on COTS hardware run into performance limitations. And he puts Marvell’s timeline “at least a year” ahead of Qualcomm’s recently announced plans for vRU and vDU chips, with availability of products from July 2021.
He also described Intel’s FlexRAN Reference platform as a “first generation product that seeded the market”.
“We provide significantly better capacity for lower power in the same PCIA card form factor, and we are providing a much more complete solution than FlexRAN does. We think it will further accelerate the market for vRAN.”
As for DU performance, with its real time processing requirements, Singh said that the model that’s been deployed by the server CPU manufacturers was to say “you can do it all in software”.
“And that runs to the limits of the real time performance you can get out of it, whether you can meet timing constraints, and that then leads to a curtailing of capacity, and you end up in a circular argument wondering why you need racks of servers to replace one box.”
“We are offering full virtualisation and containerisation such that you get the capacity and real time processing without stacking servers.”