With 5G network deployments well underway across the globe, network operators are turning their attention to getting synchronization distribution performance to the right level to support demanding 5G requirements. In this blog, the second in a series focusing on synchronization, we look at the growing importance of synchronization performance in DWDM networks. Within the industry, there is a lot of focus on G.8275.1 full on-path support and the relative Class A, B, C, or even D performance of network switches and routers, but the interconnecting DWDM network is a critical component in this discussion, as we will discuss in this blog.
Time-division duplex (TDD) networks, either 4G-LTE or 5G, require 1.5 µs maximum time error at the cell site to ensure compliant operation. The maximum absolute time error (Max|TE|) is subdivided into smaller error budgets for differing segments of the network, as shown in Figure 1 below.
Figure 1: G.8271.1 time error budget and network reference points
This allocation of time error allows for a total of 1,000 ns for the transport network between the telecom grandmaster (T-GM) and the telecom time slave clock (T-TSC) at the cell site, as shown between reference points B and C in Figure 1 above. This time error budget is divided between asymmetry in the nodes and asymmetry in the links (fibers). Managing this asymmetry is of paramount importance in building a 5G-quality mobile transport network. The remaining budget includes ±100 ns for the T-GM; ±150 ns for the end application, which is essentially the base station in a mobile network; and ±250 ns for short-term holdover in the base station to allow for switching to an alternative T-GM in failure scenarios, etc.
The primary reason that asymmetry management is so important is that 1588v2 fundamentally assumes that the network is symmetrical, with exactly the same delay in both directions. Understanding the transit time from the T-GM to the T-TSC is a critical part of 1588v2 operation, and this is determined by measuring then halving the round-trip time for a precision time protocol (PTP) packet to go from the T-GM to the T-TSC and back again. In a totally symmetrical world, this method would give an accurate calculation, but in reality, as we discuss in this blog, there are lots of sources of asymmetry in transmission networks that impact this measurement and need to be managed to enable 1588v2 operation in telecom networks.
The situation is further complicated as these time/phase errors are not static over time itself. Therefore, the Max|TE| is calculated from understanding both the constant time error (cTE) of a node, link, or network and the corresponding dynamic time error (dTE), as shown in Figure 2.
Figure 2: Maximum time error relationship to constant and dynamic time errors
Time error (TE) at any given time is the sum of cTE and dTE, as shown in green in Figure 2. Max|TE| is the maximum observed absolute value of TE in the network measured from zero; it is represented as a time, usually in ns, and is always a positive figure. cTE, shown in orange, is constant time error, which again is represented as a time figure in ns and can be either a positive or negative figure. For network components with a static error, such as optical fiber, cTE is the same at any instance in time. For network components with a more dynamic nature, such as an IP router, then the standards define that cTE is calculated using an average measurement of time error over a 1,000-second period. dTE is also represented in ns, although as it varies over time, it is usually specified as maximum time interval error (MTIE), as shown in blue in Figure 2. Max(dTE) is the maximum dTE measured from the cTE, and Min(dTE) is the minimum dTE, again measured from cTE, giving a negative value.
Looking at the mobile transport network, the main consideration in synchronization-friendly network design is managing both the constant and dynamic time errors throughout all network components, paying particular attention to the asymmetry.
Figure 3: Sources of asymmetry in optical transport networks
The main contributors to time error in optical transport networks can be summarized as follows:
Fiber asymmetry within the network. DWDM is typically unidirectional, with each fiber being used for transmission in one direction only and a fiber pair being used for a bidirectional transmission channel. Differences in the lengths of the fibers over the route will create a constant time error. Differences occur in outside plant fiber, patch cable length, repair splicing, etc. Each meter of fiber length asymmetry creates 5 ns of additional latency and 2.5 ns of corresponding cTE. This asymmetry is predominantly static but will change when fibers are repaired following fiber cuts or when patch cables are changed during network maintenance or reconfiguration.
Dispersion compensation for non-coherent DWDM. Many access networks either are not yet using coherent optics or mix coherent with 10 or 25 Gb/s on/off-keyed optics that require dispersion compensation. Dispersion compensation based on compensating fiber (DCF) is most common and uses lengths of fiber cut to meet a dispersion requirement rather than constant length. Variable length creates variable cTE issues in synchronization networks. Dispersion compensation modules (DCM) based on fiber Bragg gratings rather than fiber remove this issue, but these are less common in brownfield networks due to the higher cost.
First-in first-out (FIFO) buffers in coherent optics. DWDM optics operating at 100 Gb/s and above use coherent optics that contain FIFO buffers within the digital signal processor (DSP). These buffers have a random latency/delay upon initial startup, which varies in each optical interface and therefore varies in each direction, creating asymmetry. This creates a random time error that is constant (cTE) over the shorter term but can sometimes be dynamic (dTE) over the longer term if there are restarts on a link due to intentional network maintenance or unintentional network instances such as fiber cuts, etc. These events are not a common occurrence on an individual link in an operational network, but the size of the random cTE that can be created on initial startup and in restarts can be significant.
DWDM transponders and muxponders based on OTN mapping. OTN mapping chips also utilize FIFO buffers, which have a latency that varies on initial startup and restarts. These deep FIFO buffers are used in OTN mapping to enable the devices to accommodate a wide range of service types and can cause an even larger latency/delay than those in coherent optics. As with the FIFO buffers in coherent optics, the figures here do not vary once the network is up and running, but the size of this error is random across a large range, created on initial startup and every restart, and differs each direction.
Time error in IP routers and Ethernet switches. Asymmetry within the router/switch can be created through inaccuracies in timestamping.
Overall, these elements can be summarized as follows:
Returning to the network limits outlined in G.8271.1 and the allocation within this for node and link asymmetry, it is clear that careful design of the underlying DWDM-based transport network is required as a single pair of OTN transponders could potentially use the whole network's link asymmetry budget. The dTE elements of Max|TE| are largely generated by switching/routing devices that can be managed through the use of G.8273.2-compliant devices. The cTE elements of Max|TE| are either large static figures that can potentially be compensated for within boundary clocks or random elements from coherent optics and OTN mapping. These random cTE elements can be managed through the careful selection of optimized packet optical and DWDM devices with a significantly lower, and more acceptable, level of random cTE, or through optical timing channel techniques that can bypass these elements totally. Without the careful management of dTE and both static and random cTE across the complete end-to-end 5G transport network, these time error limits can be costly and very hard, if not impossible, to achieve.
For those readers that want to dive into this topic in more detail, our new Synchronization Distribution in 5G Transport Networks e-book provides a detailed overview of synchronization distribution challenges and standardization along with an end-to-end synchronization distribution strategy that meets the demanding requirements that 5G is driving into optical networks.