Spectrum analysis capabilities implemented in custom silicon is the recommended implementation. These hardware-based solutions speed up data collection and processing of RF information (FFTof the spectrum and RF energy bursts) as compared to software-only based solutions. This is also directly reflected in the granularity of resolution of the RF data that is presented to the user. With these capabilities, the spectrum solution is able to keep up with all the dynamic or transient changes that are occurring in the RF environment. In ASIC based solutions, it is important to remember that the firmware also plays a major role and is responsible for the actual device classification, giving the entire solution the flexibility to receive updates as new interference devices need to be added to the default classification database. If this capability were not available, hardware upgrades would be needed every time the vendor has new entries for the classification database, which in most cases would be unacceptable.

Software-based spectrum solutions using Wi-Fi adapters that are traditionally built to see only Wi-Fi signals may see certain bursts of RF energy that may not be deciphered, but can somehow still be used to calculate basic interference levels or scores. It cannot deliver the critical information to solve any real issue, which includes accurate detection of non Wi-Fi interference sources, the channels they impact, level of impact, location capabilities, etc. Software-based spectrum capabilities are used more as a feature check-box item for RFPs or websites that are used by vendors when they are asked about their RF spectrum management capabilities. Novice users may be influenced by this and accept these solutions, regretting it later when faced with the challenge of solving a real problem.

Effective spectrum analysis solution requires ASIC capabilities, including hardware accelerators and dedicated capture memory.
The variety of interference devices in the unlicensed band is broad, and is only growing over time. For this reason, a simplistic and small set of classifiers is not sufficient. Software-only spectrum analysis solutions do not have sufficient capability to scale to a large set of classifiers and operate effectively.

The first reason software solutions can’t scale is that the total horsepower (MIPS/capture-memory) required grows with the number of classifiers. Think of each classifier as a theory that must be tested vs. an unknown energy source. In other words, when we see energy, we must test: Is it Bluetooth? Is it a cordless phone? Is it a jammer? Is it WiMax? This causes the amount of processing to grow with the *number* of classifiers.

The second reason software solutions can’t scale is that when you start to add more classifiers you need to change the way you perform the analysis. With a simple small set, it’s possible to look only at the “fingerprint” of the interference – the timing of pulses and center frequency. But when you add more classifiers and need to separate between devices that are similar from an RF perspective (or to detect how many devices are actually present), you need to do deep “DNA” analysis of the signals. This takes a tremendous amount of DSP processing power.

When a software-only solution runs up against the wall in terms of MIPS and/or capture-memory, it starts to do one of two things: A) it does not run all the tests, or B) it cuts back on the quality of analysis it runs before declaring a particular device is present. Not running all the tests results in non-detects – ie. missing devices. Cutting back on the quality of analysis results in false detects – seeing ghosts. Both of these affects are intensified in a busy RF environment, where there is a lot of activity, and it’s very easy to confuse normal activity with interference.

Non-detects are a big issue. When a user calls in with a wireless problem, IT wants to diagnose the source of the problem. The first thing IT wants to determine is whether the problem is at the physical layer. So, Spectrum Intelligence is supposed to reveal whether the physical layer is OK or not. But if the system is subject to non-detects, it can't be trusted. The system reports nothing, but IT is still left with the nagging feeling that interference might be the issue. In other words, the use case has not been solved.

False detects are also a big problem. In the end, the model of a customer calling with a problem (reactive model) is undesirable. What IT wants is a system that can be more proactive, and alert about interference before it becomes a problem. But a system that false detects is one that sends IT on a wild goose chase. Every time there is a false detect, IT wastes time chasing down a ghost. IT must trust that the system does not "cry wolf", or they will simply turn off pro-active alerts.

To summarize, the only effective spectrum analysis solution that can scale to a large set of classifiers, without false detects and non-detects is a system with sufficient ASIC and memory horsepower.

On the question on ASIC vs. software, both seem to agree strongly. So does anybody do this via software?

What's the advantage of software? Easier upgrades?

And if the ASIC is programmable, doesn't this imply "software"?

So doesn't the ASIC lead to a more extensive/expensive upgrade?

There is no advantage to a software-only solution. Some WiFi vendors are limited by the capabilities of their WiFi chipsets, and so are trying to make do with what they have. But the performance of these software-only solutions will not meet Enterprise requirements.

In terms of upgrade, the Cisco solution consists of both ASIC HW and software. The software portion is upgradeable to add more classifiers over time (as new types of unlicensed band devices are introduced). There is no cost to these upgrades.