Cross-Connect
How a Cross-Connect Grooms and Routes Transport Traffic
A cross-connect sits at the hub of a transport network, where multiple aggregate links converge and traffic must be rearranged before it continues toward its destination. Unlike a router that examines packet headers, a cross-connect operates below the service layer: it manipulates fixed containers, such as SDH virtual containers, OTN ODUk frames, or optical wavelengths, and simply maps an input container to an output container. Because the mapping is provisioned rather than per-packet, the fabric is deterministic, introduces negligible jitter, and adds only the fixed latency of the switch matrix, typically a few microseconds for an electrical fabric.
The two dominant implementations are the digital cross-connect (DXC) and the optical cross-connect (OXC). A DXC demaps each line signal into its constituent tributaries, switches them through an electrical fabric, and remaps them onto outgoing lines, giving fine grooming control at the VC-12 (2 Mb/s) or VC-4 (140 Mb/s) level. An OXC keeps traffic in the optical domain, steering entire wavelengths through a MEMS or wavelength-selective-switch matrix, which scales to terabits but cannot groom below a single wavelength. Carrier networks frequently stack both: an OXC handles bulk wavelength routing while a DXC grooms sub-wavelength tributaries onto those wavelengths.
For millimeter-wave backhaul, the cross-connect is the point where many capacity-limited radio hops are consolidated. By grooming several partially filled links into fewer fully loaded ones, the operator reduces the number of hops and spectrum it must license, and the same fabric provides the sub-50 ms protection switching that keeps a ring or mesh resilient when a single hop fades.
Non-Blocking Fabric Architectures
The internal switch matrix determines how many simultaneous connections the cross-connect can support and whether it ever has to disturb an existing path. A full crossbar of N inputs by N outputs is strictly non-blocking but scales as N squared crosspoints, which is impractical above a few hundred ports. Large cross-connects therefore use a three-stage Clos network, which reaches the same connectivity with far fewer crosspoints at the cost of occasional rearrangement of in-service paths.
Cross-Connect Sizing Equations
Crosspoints = N × N (N = number of ports)
Clos rearrangeably non-blocking (3-stage, n inputs per module):
Crosspoints ≈ 2N√(2N) when m = n, with N = n × k
Strictly non-blocking Clos (Clos condition):
m ≥ 2n − 1 (m = middle-stage modules, n = inputs per ingress module)
Grooming fill efficiency:
η = ∑(used tributaries) ÷ (links × tributaries per link)
Example: an N = 512 port DXC built as a Clos network needs roughly 2 × 512 × √1024 ≈ 33,000 crosspoints versus 262,144 for a full crossbar, a 7.9× reduction.
Cross-Connect Type Comparison
| Attribute | Digital XC (DXC) | Optical XC (OXC) | Patch Panel (manual) |
|---|---|---|---|
| Switching domain | Electrical timeslots | Wavelengths / fibers | Physical fiber/copper |
| Granularity | VC-12 / VC-4 / ODUk | 1 λ (~100 GHz or flexgrid) | Whole port |
| Capacity (typical) | 40 to 640 Gb/s fabric | 1 to 100+ Tb/s | Limited by ports |
| Reconfig time | < 50 ms (protection) | 5 to 100 ms (MEMS/WSS) | Minutes (manual) |
| Grooming | Sub-wavelength, fine | Wavelength only | None |
| Insertion impact | O/E/O regeneration | 0.5 to 6 dB optical loss | 0.2 to 0.5 dB per mate |
| Best application | Backhaul grooming, TDM | Core wavelength routing | Static rack patching |
Frequently Asked Questions
What is the difference between a cross-connect and an add-drop multiplexer?
An add-drop multiplexer sits on a ring or chain with two line ports and a set of tributary ports; it drops and adds local channels while pass-through traffic stays in place. A cross-connect has many aggregate ports and a switch fabric that maps any incoming channel to any outgoing channel, enabling full any-to-any grooming, mesh interconnection, and ring-to-ring hand-off. A DXC at VC-12 or VC-4 granularity rearranges thousands of channels, whereas an ADM typically serves a single ring. Modern packet-optical nodes integrate both functions in one shelf.
What does grooming mean in a digital cross-connect?
Grooming consolidates partially filled tributary streams so high-capacity links carry full payloads instead of stranded fragments. Several STM-1 feeds each at 40% utilization can be groomed into fewer fully loaded STM-1 or STM-4 channels, freeing capacity on the backhaul radio. A VC-12 fabric grooms at the E1 (2 Mb/s) level; a VC-4 fabric grooms at 140 Mb/s. Effective grooming cuts the number of leased links and microwave hops, lowering operating cost and raising fill efficiency on capacity-constrained millimeter-wave backhaul.
What is a non-blocking cross-connect fabric?
A non-blocking fabric guarantees that any free input can always reach any free output, so no admissible request is refused for want of an internal path. A strictly non-blocking crossbar of N by N needs N² crosspoints and never reroutes existing paths. A rearrangeably non-blocking Clos network reaches the same connectivity with far fewer crosspoints (on the order of N¹∙⁵) but may reroute in-service connections to admit a new one. Carrier-grade units are specified as strictly non-blocking, or rearrangeably non-blocking with hitless rearrangement.