System and components models are a very good thing. They let designers simulate before actually building, test tradeoffs, and more. They can install a level of confidence that helps designers go to the next step. But good as they may be, they are not perfect for many reasons; no need o go into them here–you all know what they are. That's why it is important to use a hardware channel simulation when it is appropriate, so you can test real-world performance of actual circuits and systems.
You can get RF channel simulators from broad-line, major test and measurement equipment vendors as well as smaller, specialized companies who focus on a more limited set of link types. While such a "box" is not necessarily a perfect model of the actual channel, it can be very close to it, and in some ways it is better: it's repeatable, measurable, and quantifiable. Whether you are doing a basic wireless link such as between a base station and a handset, or a GPS receiver with its downlink-only link, or a satellite system with uplink and downlink, a hardware simulator can provide the harsh reality you need.
Several years ago, I met with a vendor of GPS-receiver IC chipsets. They also provided a reference design with full receiver front-end circuitry (schematic, layout, and BOM). I asked them how they tested their GPS core engine, and they said they used a set of industry-standard test vectors. That seemed OK enough. Then I asked them how they tested the "ready to go" reference design and the answer was, in essence, "we stuck an antenna out the window at our lab, and it seemed to work that tme."
Well, that's certainly reassuring, right?. All it proves is that on that one day, under one mostly unknown set of circumstances, the GPS receiver worked to some extent. But as all engineers now, it's pretty risky–yet all too easy–to extrapolate from as single point. And before you know it, the "it seemed to work, once" gets translated and codified as "sure, it works just fine." ♦