Primer on Implementing VoCable Architectures: Part 2

Primer on Implementing VoCable Architectures: Part 2
In the first part of this article, we discussed the fundamental elements of a voice-over-cable (VoCable) system (See Primer on Implementing VoCable Architectures: Part 1) . We will now explore the attributes necessary to successfully integrate these elements into a viable, quality telephony service infrastructure.

Quality of service (QoS) is essential to successful cable telephony deployment. While consumers may choose to accept less than toll-quality services to obtain other features (mobility for example, as illustrated in the wireless telephony market), they are not expected to tolerate inferior quality for home and small business phone calls.

Though circuit-switched telephony services based over cable networks are deployed today, the use of IP-based solutions afford opportunities for other QoS applications. But, as part 1 pointed out, the cable network presents issues unique to IP Telephony services deployment. To maintain appropriate QoS, the limited, asymmetrical bandwidth of the cable network resources requires a prioritization and permission scheme for the customer premise equipment (CPE) transmission of data and voice packets. This important factor has been addressed in the data over cable service interface specification (DOCSIS) standard and technology for IP-based cable telephony.

DOCSIS 1.1 establishes four primary service categories: unsolicited grant service (UGS), real-time polling service (rtPS), non-real-time polling service (nrtPS), and best effort (BE) service. In a UGS flow, the cable modem is guaranteed to receive fixed size grants at periodic intervals from the cable modem termination system (CMTS), without the need to explicitly send requests. Let's look at UGS and nrtPS.

The use of a UGS reduces latency by eliminating the need to request a grant cycle for every packet. A tolerated grant jitter is negotiated at service setup, in addition to the grant size and the period.

While it provides reduced latency, UGS is not the best solution for applications that don't require a constant data rate. For example, it is inefficient to use UGS for applications (such as voice with silence detection) that don't require a constant data rate over time. A better option in these situations is a variant of UGS, dubbed the unsolicited grant service with activity detection (UGS-AD), which enables the CMTS to detect flow inactivity (i.e. lack of grants used by the cable modem), by sending unicast request opportunities (also called "polls") at periodic time intervals. The cable modem can use the unicast requests opportunities to send requests to resume voice transmission avoiding latency incurred by contention.

In a nrtPS flow, the cable modem is guaranteed to receive unicast request opportunities at periodic intervals from the CMTS. If the cable modem does not use the request opportunities, the CMTS allocates the reserved bandwidth to other flows, mitigating the inefficiency of UGS. In a nrtPS flow, the bandwidth is not guaranteed to the flow, however, the cable modem is allowed to use multicast request opportunities for the flow.

More on flows

The flow in which a packet is transmitted is based on the content of the IP header fields, allowing every application to receive a different service flow. Multiple data flows (each flow corresponding to a service and identified by a service identification descriptor [SID]) concurrently exist in a cable modem. A transmission request in the upstream and the corresponding grant includes the SID as the flow identifier. The cable modem and the CMTS negotiate the QoS for each flow upon allocation and dynamically as the service requirement changes. QoS is then achieved by the implementation of sophisticated scheduling mechanisms in the CMTS. A classification function is applied to every packet.

The PacketCable dynamic QoS (DqoS) specification applies to the voice and other applications layers. DQoS utilizes various protocols, and leverages DOCSIS 1.1 to allow for provisioning, transport and billing of varying levels of service classifications. A bi-directional data flow session is established between two clients. By assuring that cable modem and CMTS adheres to the QoS rules, the DQoS spec provides for the necessary resources to be reserved and then subsequently committed for each data (voice) flow session. Two-phase (reserve and commit later) and single-phase (commit) QoS activation models are required, as well as resource changes during a session and dynamic binding of resources (e.g. transferring call session parameters for call waiting).

3 Key Attributes

Three significant network attributes must be properly addressed for acceptable voice quality in a packet-based cable network.

1. Packet loss, which may occur in a high congestion or high latency situation, has a direct effect on voice quality so it is imperative that some type of packet loss concealment technique is employed.

2. Delay or latency of packet transmission is another issue that affects the quality of IP telephony. At approximately more than 200ms of delay, the latency is perceptible to the listener and doubletalk may take place.

3. Due to the asynchronous behavior of the packet network, the inter-arrival rate of the packets can vary. This jitter can mean that the next packet to be transmitted to the user on the receive side may not arrive in time.

All of the issues above contribute to problems in a voice transmission, which can be deadly in a VoCable architecture. Several techniques are available to mitigate these problems. First, some vocoder algorithms include packet loss concealment techniques. For single packets lost in a sequence, a simple method of replaying the last packet in place of the lost packet hides this loss.

Redundancy schemes, i.e. sending multiple versions of each packet, can also be used. But, these schemes have a significant impact on bandwidth utilization, and therefore are not common for voice transmissions. Redundancy schemes are, however, often used to assure accurate transmission in fax relay connections. The control packets of a fax call are transmitted in duplicate or triplicate since a lost packet would result in fax connection failure.

A playout buffer in the VoCable gateway can also help curb effects of latency. Clearly, slowdowns in the VoIP gateway and packet loss in a network can contribute to end-to-end latency. For example, accumulation delay, which is the amount of time necessary to collect a frame of voice samples to be processed by the vocoder, can occur in the gateway. This can be 2.5ms to 10 ms for typical vocoders used in VoCable applications. The playout buffer can curb delay problems, by the gateway and the network, by queuing several packets before beginning voice playout to the listener.

A jitter buffer, often the same as the playout buffer just described, is also necessary to stop latency. However, if the buffer is improperly sized, packet loss, or additional latency may be induced. It is ideal that the jitter buffer be adaptive to the dynamic packet loss and network jitter of the network at any given point in time. This requires real-time statistic collection along with the ability to automatically adjust the buffer size accordingly.

Bandwidth minimization

Although not a direct contributor to the quality of a given voice connection, bandwidth minimization is also important to allow the maximum possible amount of voice connections and data transmission over the cable network especially as consumer take rates increase. The use of low bit rate (LBR) codecs greatly reduces bandwidth utilization.

Additionally, the PacketCable codec specification includes the use of a low delay vocoder (G.728) and a high fidelity vocoder (G.729E) when using compression of voice, further improving voice quality. In the future, as VoCable systems proliferate and where IP connectivity is end-to-end, higher quality vocoders, such as G.722 may be used to exceed toll quality.

The quality of the echo canceller implementation can also play a role in the quality of the cable IP telephony call. Compliance with G.168 2000 performance measurement standard is a minimum requirement. The echo canceller should be able to detect a double-talk condition and assure cancellation of only the echo itself, not one of the speakers. It should also have robust performance in a high background noise situation, a condition which can cause some echo canceller processing to be derailed.

In addition to meeting all of the necessary standards for a voice-enabled cable modem, the components of the solution must work effectively as a system to achieve standard execution in a efficient, yet cost effective solution. The synchronization of voice frame packets, starting with the grant from the cable network, can further reduce processing delay in the cable modem. Providing an efficient path for the voice packets (UDP packets) through the software and hardware stack within the cable modem can offer additional processing delay reduction.

System-level view

Now that we've looked at some techniques for improving QoS, let turn to a system level view of a VoCable system that integrates the elements and attributes as previously described. The suite of PacketCable standards defines a system for QoS based IP services (see Figure 1) with telephony expected to be one of the first implementations. PacketCable 1.0 defines the requirements for a single zone telephony system, PacketCable 1.1 primary line requirements, and PacketCable 1.2, inter-zone specifications.

Vendors working with CableLabs are aggressively testing standards-based equipment at DOCSIS and PacketCable 1.0 certification sessions, assuring equipment compliance and interoperability. In the meantime, multiple trials are taking place around the world, using a mix of proprietary schemes and some of the CableLabs standards.

An Interim solution

To reduce risk and capital investment of migration from traditional telephony and circuit-switched cable to a pure IP-based network, an interim solution has been proposed. As part of an additional PacketCable activity, a technical report (PKT-TR-ARCH-LCS-V01-010730) describes the use of an IP data terminal (IPDT) that receives the IP packets as forwarded by the CMTS, and translates voice to existing telephony switch equipment via a GR-303 interface. This provides a packet-switched based system without converting to a pure IP network. The intent is that the IPDT equipment can migrate to media and signaling gateway implementations (see figure 1).

Hybrid fiber coax (HFC) networks offer a viable infrastructure for VoCable service opportunities. These networks leverage field-proven VoIP technology to increase deployment success and reduce time to market. As a result of their initial VoCable trials, multiple system operators (MSOs) have declared that the fundamental technology required for VoCable deployment works.

One key aspects to a successful VoCable deployment is meeting or exceeding the current quality of service experienced by PSTN telephony services today. The use of IP technology versus traditional circuit-switched technology offers opportunities for telephony services that exceed today's PSTN capabilities. Voice encoding schemes can offer higher than toll quality clarity. Additionally, features such as customized announcements, ringing cadences, and unified messaging can be efficiently offered by the MSOs.

Debbie Greenstreet is the product management director for Telogy Networks, a Texas Instruments company. She holds a BSEE from the University of Virginia and has done graduate work in computer engineering. Debbie can be reached at dgreenstreet@telogy.com.