A "Budget" Spectrum Analyzer Cost Me $12,600 — The TCO Lesson That Led Us to Keysight

If you've never had to explain to your boss why the brand-new, budget-approved test equipment can't actually do the test you were hired to do — well, you're either very new to this industry or very lucky. I was neither.

This story starts in March 2022, when our small pre-compliance lab in Austin took on a project that should have been straightforward. A networking startup wanted us to pre-test their new 100G Ethernet switch — we'll call it the C210 — against a Cisco Catalyst reference unit. They were preparing to pitch the C210 to data-center customers, and their engineering director wanted to know two things: would it pass FCC Part 15 Class A, and how did its radiated emissions compare to the Cisco switch?

That comparison was their idea, not mine. Comparing your product to an industry standard is risky, but if the data comes out looking good, it's marketing gold. My job was to ensure the data was trustworthy. This is the story of how I almost failed at that — and why the lesson stuck with me.

The Budget Decision That Made Me Look Smart (Until It Didn't)

We'd just received a test-equipment budget of $15,000. Up to that point, our lab had been limping along with a used oscilloscope and a spectrum analyzer older than I am. My boss told me to research and recommend what we actually needed. Notice he didn't say "the cheapest thing that technically works." I decided that part myself.

After a few days of comparing spec sheets, I placed the order: a "value-priced" spectrum analyzer covering up to 1 GHz, and an RF signal generator that could go to 3 GHz. Together, they came to about $4,300. It felt like a no-brainer. I remember telling my senior colleague, "We're not buying big-brand gear just for the logo. We're buying what the job actually needs."

If you've ever said those words out loud, you know the bell should go off right there.

The job needed two things: an accurate measurement of emissions from the C210 while it was running a real traffic load, and a calibrated RF source to test immunity to external interference. My budget gear promised both. In practice, it did neither.

Day One: The Noise Was Everywhere

The first red flag appeared within an hour of setting up. I had the C210 on the test bench, connected to a Cisco Catalyst switch that was generating the traffic load — a 100G link between them through a long cable that I'd routed, for convenience, right past my measurement antenna. In my defense, the lab wasn't a proper anechoic chamber, and routing the cable the other way would have meant a longer patch cable. In hindsight, that convenience cost us the entire first day.

The spectrum analyzer showed a noise floor like a mountain range. Everywhere I looked, there was energy. I couldn't tell actual emissions from ambient noise. I tweaked settings, swapped cables, tried to compensate — the picture stayed muddy. For far too long, my conclusion was "the C210 has serious noise issues." Actually, I was measuring a combination of the Cisco switch's own power-supply noise, the poorly placed cable acting as an antenna, and the analyzer's mediocre noise floor. A three-layer cake of bad decisions.

What "Switches vs Cisco" Actually Taught Me

Here's the part I wish someone had told me months earlier: when you're comparing a device under test against a Cisco switch, the reference switch is part of the test environment, not part of the measurement. Its fans, PHY clocks, and switching power supply all radiate. That's true of every switch — including the C210, which was the whole point. But if you sit one next to the other and share cable paths, you will not be able to tell which switch is emitting what.

By the end of day one, the client's project engineer asked, "So, did we pass?" And I had to say the sentence no test engineer wants to say: "I'm not sure yet. The data is inconclusive." Not a great look.

Day two, I moved the Cisco Catalyst to the far side of the lab, built an isolated test harness with proper shielded cables, and restarted the scan. The noise floor got cleaner. Not perfect, but cleaner. And that's when the second problem became undeniable.

The Signal Generator That Couldn't Make Up Its Mind

One of the immunity checks our client requested was simple: inject a CW tone at a few key frequencies — including 2.4 GHz, to simulate Wi-Fi interference — and confirm the C210's management interface stayed reachable. The point was to prove the switch would keep answering SNMP queries while sitting next to a wireless access point.

My RF signal generator could reach those frequencies on paper. But when I actually set it to 2.400 GHz and measured the output, the tone drifted by nearly 15 MHz over ten minutes. Fifteen megahertz. For a Wi-Fi immunity test, that's a deal-breaker — you'd be testing the switch at a frequency where nothing real-world ever transmits, and you'd miss the band edges entirely.

I checked the cables. I checked the grounding. I even checked the room temperature (74°F, normal). The generator was simply not stable. Honestly, I'm not 100% sure whether it was a design flaw or a defect on that particular unit — either way, it was unusable for the purpose I'd bought it for.

Calculating TCO on a Whiteboard

At that point, we were looking at a 48-hour delay and a client who was losing confidence. My senior colleague — twenty years of EMC experience — came over and asked three questions.

First: "How much did the two pieces of gear cost us?" "About $4,300," I said. Second: "How many hours of engineering time have we burned on questionable data?" I didn't want to do the math. Third: "If the client walks, what does that cost us?" I definitely didn't want to answer that one.

He wrote it all out on the whiteboard while I watched:

• $4,300 for the budget gear.
• About $3,600 in labor across two engineers for the failed first two days.
• $2,800 in client credits we used to rebuild trust after the delays.
• An estimated $1,900 in risk if the client had pulled the project and gone elsewhere.

That's $12,600. The cost difference between our budget package and a proper Keysight setup — a CXA-series spectrum analyzer plus an MXG-series RF signal generator — was roughly $6,000 at that time. In other words, being cheap was already twice as expensive as being right.

Now, to be fair, the budget analyzer wasn't complete garbage. For basic sub-1 GHz troubleshooting, it was acceptable. But the job wasn't "basic sub-1 GHz troubleshooting." The job was pre-certifying a 100G switch. Pre-certification isn't the place to hope.

When the Keysight Spectrum Analyzer Turned on the Lights

The nearby Keysight office had a loaner program, and that turned out to be our lifeline. Two days later, we had a Keysight spectrum analyzer — a CXA N9000B — on our bench, along with an MXG signal generator. I still remember the first scan.

Same C210. Same traffic load. Same test harness. But the noise floor was flat. Calm, predictable, and about 15 dB lower than what my budget analyzer had shown. It was like someone had cleaned a dirty window. Peaks that had been buried in the mud were suddenly visible.

One peak got my full attention: a narrow spur at 3.15 GHz, about 22 dB above the noise floor. It wasn't there on the Cisco switch. It was only on the C210. And it was well above the 1 GHz limit of my original analyzer — which is exactly why I'd never have caught it. Per the FCC's rules on unintentional radiators (47 CFR Part 15), emissions from digital devices like switches need to be measured across the frequency range they actually generate, and for 100G Ethernet, that extends well beyond 1 GHz. My cheap analyzer had been blind in precisely the range that mattered.

We pulled out the lab's Keysight Infiniium oscilloscope and probed the PHY reference clock inside the C210. There it was: a 156.25 MHz clock with about 11 picoseconds of period jitter — far too high. A slightly wrong termination resistor on a clock buffer was causing a low-level oscillation that showed up as that 3.15 GHz spur, the 20th harmonic. The client changed a single component value. Two weeks later, the C210 sailed through pre-compliance and passed at an accredited lab.

One aside worth mentioning: we also discovered that Keysight RF signal generators — the MXG we used on that project in particular — have phase stability that makes a joke of my old unit's drift. It wasn't until I saw a properly stable tone on the CXA's display that I realized what I'd been missing.

Bottom Line: The TCO Lesson

So what's the actual moral of this story? It's not "buy Keysight because it's the best." It's not "cheap instruments are always useless." It's about what you're really paying for when you acquire test equipment.

For me, total cost of ownership means answering three questions before any purchase: Can this instrument see what I need to see, reliably, when I need it? Can I defend the data if a client's product launch depends on it? And if something goes wrong, will the vendor support me, or am I on my own?

The bottom line: we spent $12,600 on a lesson that could have cost $6,000. The Keysight spectrum analyzer and RF signal generator weren't luxury items. They were the cheapest option that would actually do the job. And for the first week of testing, the CXA's dynamic range felt like an infinity pool next to my old gear's murky pond — the kind of confidence you can't put on a spec sheet.

If you're building a test bench for network switch pre-compliance — whether you're measuring a Cisco Catalyst, a generic 10G switch, or your own design like the C210 — don't learn this the hard way. Calculate the TCO, not the price tag. And if a senior engineer offers to show you the math on a whiteboard, you should probably listen.