GK is an optional logical entity in an H323 network. It provides address resolution, registration and status.
In MGCP, the MGC or call agent is the controlling entity of the solution. It tells the endpoint what to do and how to do it.
In 323 the endpoints ask the gatekeeper what to do. In MGCP the endpoints are told by the MGC/CA what to do.
Not everyone uses as separate signaling gateway, some do. I have seen a few call agents that use an “Ulticom” card to term1nate layers MTP layers 1-3 and then parse the various parameters of the ISUP message to create the call.
If there is a “signaling gateway” then most folks use a proprietary method to encapsulate and send uppper layer messages, sometimes those SGs term1nate MTP layers 1-3 sometimes only 1-2 and pass the rest up to a more intelligent entity who can parse the message.
The ultimate answer is today its primarily proprietary to connect a SG to a network. Rule of thumb:
In an MGCP environment the SG always talks to the CA or MGC. The signaling gateway can use the sigtran protocols to transport that information to the CA or it can use a proprietary mechanism. The reality of the situation is that SCTP, M3UA, M2UA, SUA and all the sigtran methods are still in the development phase for vendors. Even if they were already built, the intelligent endpoint would still have to understand the SCTP encapsulation as well as have the ability to parse M3UA messages and interwork that into one of the signaling/control protocols.
To connect these two networks: MGCP and H323. The call agent would have to “speak” H323. H323 recieved would be translated to an MGCP message to the mgcp endpoint. The same thing when the ingress is an MGCP network.