In the near past, home infrastructure consisted of different technologies. We needed a phone, an internet connection, a TV, and several separate corresponding devices to support these various services. These devices used to range from a DSL modem to a router, a DECT phone base, and a cable TV box. On the other hand, the current home Services portfolio includes an enhanced Cloud-Based Service Platform with a centralized, cloud-based service architecture. Service “intelligence” is implemented in the cloud back-end systems. However, at this time, home GW only seems useful for providing transparent internet connectivity.
5G Cloud-based and Fog architecture
The benefits of a cloud-centered architecture include scalability, reliability, and flexibility. A good example is pay-as-you-go services. The disadvantages, however, are that the spatial and temporal locality of service access is ignored. All service invocation is forced via a centralized service API. This means that raw data is pushed for storage and processing in the cloud, which may have a negative impact on latency, privacy, traffic, and energy efficiency. This is where fog computing can come in handy. Fog architecture may provide an end-to-end approach encompassing the cloud but penetrates also down to the very end-device. A fog model can offer dynamic offloading of the cloud functions onto an intermediate layer of more capable processing and communication devices.
At the same time, 5G technology with its ultra-high capacity and very low latency can provide the network infrastructure for the underlying connection to the cloud but also the edge. A 5G Home GW solution can play a crucial role in the emerging fog model to re-imagine the home GW as a powerful and flexible local service platform. The emerging 5G Home Services are expected to be “Intelligent” as they interact with users using speech and can be aware of their environment. This is due to their central point of control for sensors, actuators, and media.
Machine Learning
There is an expectation for explosive growth in machine learning-based services with impressive results. Speech recognition and generation, as well as language translation and sound recognition achieved, can be almost at a human level. At the same time, image classification and lip-reading can be reached at a scale that is a lot better than the current human level.
Machine learning services play an increasingly important role in the home infrastructure and services. User interfaces such as conversational UI and automation practices such as self-learning control algorithms can support this notion. Furthermore, face and person recognition, together with sound recognition functionalities, can enhance overall system security.
Cloud-based ML-Service Model
Machine learning needs considerable resources, a lot of training data, and a lot of processing power. Current home machine learning services follow a cloud-centered approach where model training and run-time inference are in the cloud. End-devices like Amazon Echo and Google Home send raw data to the cloud and receive results back.
On the other hand, a Fog-based ML-Service Model can increase privacy and reduce latencies. These can be achieved by performing these machine learning tasks locally. However, machine learning on the Home GW includes some challenges such as how to offload some of this work and boost the processing resources on the GW. An example could be Tesla installing NVIDIA Drive PX 2 “deep-learning supercomputer” in each car. The idea is to provide resource-efficient training and inference on resource-constrained HW.
Machine Learning on Home GW
The key idea is to trade-off inference accuracy for a reduced resource footprint. Machine learning service components can run on the home GW. This can be achieved by using pre-trained models that are only customized locally with additional data. Furthermore, compression of pre-trained models will be able to execute ML-based services on resource-constrained HW.
Distributed Home Services
The hybrid Fog model can be used as an intermediate step to a distributed peer-to-peer service architecture. This solution will offer a flexible migration of the functionality across different layers of the architecture. Furthermore, it could offer opportunistic matchmaking between the providers and consumers of services. Distributed Home Services can enable autonomous service transactions without human participation. For example, the washing machine or refrigerator could manage their own supplies and schedule to call a technician when broken.
DHS Key Enabler: Blockchains
Blockchains offer a permanent distributed record of interactions between individual actors in a decentralized system. A distributed public ledger can keep a chain of persistent records protected against tampering and revise using cryptographic mechanisms.
Ethereum Smart Contracts
Ethereum supports storing rudimentary programs and their results on the blockchain, so-called smart contracts. This is the core technology for many pilot projects and has received major backing from Microsoft in tooling on Microsoft Azure.
DHS Blockchain Challenges
A full node can verify the current state of the blockchain depending only on local resources under its control. Furthermore, consensus algorithms require significant resources such as processing for “proof of work” or monetary for “proof of stake”. Both are out of reach for a typical home GW. Home devices currently use a client/server model with a remote procedure call (RPC) to a full external node to initiate and verify transactions.
Enabling Blockchain on Home GW
The key idea is operating on a partial view of the blockchain. Blockchains use the concept of Merkle Trees. Merkle Trees allow efficient proof that a given key/value pair is indeed stored in a tree with a particular root hash. A separate protocol is needed to download block headers and Merkle proofs on demand from a full node, for example, the Light Ethereum Subprotocol. This offers a high-security assurance about the current state of a part of the blockchain state, such as locally verifying the execution of a transaction).
Home GW Requirements
Home Gateway requires enlarged computational capabilities such as generic multi-core CPUs, dedicated GPU, FPGA, ML, and security accelerators. Moreover, extended communication capabilities such as WLAN, BLE, and Thread / 802.15.4 are needed. Regarding software requirements, a Home GW needs software support for isolation of the multitude of service components running on the home GW. Furthermore, it requires support for flexible loading/unloading, discovery, and management of these components, potentially by external service providers. The limited solution is offered by isolation on the application level (e.g. Eclipse Kura, OpenHAB following the OSGi plugin model).
There is also a need for full virtualization and containerization of the individual software components. This includes thin base OS images and native support for transactional software updates and binary patching, such as Snappy Ubuntu Core. Lightweight containers such as docker on ARM and snaps should also be included, together with container orchestration tools (e.g. docker swarm, Kubernetes).
Cooperating 5G Home GWs in the fog can offer a robust and flexible service platform. This can be achieved by supporting novel communication services and utilizing extended context information and machine learning intelligence. Further advantages include a distributed WLAN radio resource management, sharing of neighborhood WLAN infrastructure, pre-caching videos, and complex event notification. With all the above advantages and the upcoming technological trends, 5G Home GW service and context management can be a major future business opportunity for Operators worldwide.
