How do I verify the output power and wavelength accuracy on a used Keysight 8168F?

Use a calibrated optical power meter (Keysight 8163B with 81623B sensor, or equivalent) to measure output power at the front panel connector. Compare against the spec card: the 8168F is rated +7 dBm typical output with power accuracy of ±0.5 dB. Wavelength accuracy requires a wavemeter — the HP 86120C or Agilent 86142B spectrum analyzer can resolve it to ±0.001 nm. If you do not have a wavemeter, a DWDM channel plan checker will confirm center wavelength to ±0.05 nm. The most common failure on used 8168F units is a degraded output power that traces to aging of the DFB laser chip inside the module. If output is more than 1.5 dB below spec and the front panel shows no error, measure through the internal 8168A mainframe power monitor — if they disagree, it is the output connector or internal coupler, not the laser chip itself.

Optical Sources Published July 21, 2026 · Aumictech Labs

O-band laser sources: what they are, which one you actually need, and why everyone suddenly wants them

Q: What is the difference between the Keysight 81600B and the 8168F for O-band testing?

The 81600B is a mainframe-based tunable laser in the 816x family that covers both C and L band; it does not natively cover O-band without the optional 81618A O-band module, which is rare on the used market. The 8168F is a dedicated O-band tunable laser source designed specifically for the 1260-1360 nm range. If you need O-band specifically, the 8168F or its predecessors (8167A fixed, 8168A/C tunable) are the cleaner choices and easier to source. The 81600B is the right tool when you need the same hardware to cover C-band for coherent testing and occasionally dip into O-band — but that is an uncommon requirement outside of vendor labs building optical amplifier test benches.

Q: Is the Anritsu MG9638A good for O-band testing?

It is a solid unit for the application. The MG9638A covers 1260-1360 nm (O-band) with a linewidth of less than 1 MHz and output power up to +6 dBm. Its step resolution is 0.001 nm, which is fine for CWDM channel testing. The main limitation is that it does not cover C-band, so if your test plan also includes 1550 nm testing you need a separate source. Anritsu's older MG9638A and the MG9639A are well-built and parts are still available through Anritsu service. On the used market, the most common issue is the front panel APC connector being scratched or the key lock being stuck — neither is a laser problem, both are cheap fixes. We verify linewidth with a self-homodyne measurement before shipping any MG9638A.

We sell and test O-band sources regularly enough that the questions keep repeating. The hardware is not complicated, but the used market has enough gotchas that it is worth explaining what to check, which models are actually worth buying, and why the demand spiked so sharply in 2023-2024.

What O-band actually means

O-band stands for Original band. It covers 1260 to 1360 nm and predates C-band (1530-1565 nm) and L-band (1565-1625 nm) as the first commercially deployed fiber optic wavelength range. The reason it got picked first is simple: standard single-mode fiber (ITU-T G.652, Corning SMF-28 and its successors) has its zero-dispersion wavelength right around 1310 nm. Zero dispersion means pulses do not broaden as they travel, which lets you run high data rates without dispersion compensation hardware. That property made 1310 nm the wavelength of choice for early Ethernet over fiber (1000BASE-LX) and SONET OC-48 links.

The tradeoff is that silica fiber has higher attenuation at 1310 nm than at 1550 nm — roughly 0.35 dB/km versus 0.2 dB/km. So for long-haul links over 80 km, C-band won. But for anything under 40 km, and especially inside datacenters where you want no dispersion headaches and you can tolerate slightly higher loss, O-band is exactly the right tool.

Why everyone suddenly wants one again

For about a decade after SONET networks matured, O-band test equipment sat mostly idle. Labs that had a Keysight 8168F tunable laser were using it for C-band transceiver work or letting it collect dust. That changed fast starting in 2022 and accelerated through 2023-2024.

Two things drove it. First, 400G datacenter interconnect settled on O-band CWDM channels as the physical layer standard. The 400G-DR4 MSA uses four parallel single-mode fibers each carrying one 100G lane. The 400G-FR4 and 400G-LR4 pack all four channels onto one fiber using CWDM wavelengths at 1271, 1291, 1311, and 1331 nm. Every hyperscaler qualifying these transceivers needs a way to characterize each of those four channels independently — which means tunable O-band sources, O-band power sensors, and O-band optical spectrum analyzers in quantity.

Second, silicon photonics foundries ramped hard to supply those 400G transceiiver chips. A silicon photonics chip designed for O-band has grating couplers, waveguides, and ring resonators all optimized for 1310 nm. Characterizing them at wafer level requires a tunable O-band source with clean linewidth and flat power across the band. Labs that had been doing C-band coherent research found themselves needing O-band hardware they had never bought before. That drove used prices up and availability down, and it is the main reason we get multiple O-band source inquiries per week now versus maybe one or two per quarter previously.

Fixed vs tunable: which do you actually need

This is where people spend money they do not need to. A tunable laser that sweeps 1260-1360 nm costs considerably more than a fixed-wavelength 1310 nm source. The question is whether you genuinely need the tuning.

You need a tunable source if:

You are characterizing a CWDM multiplexer, demux, or filter across the O-band and need to map the passband shape
You are testing a DFB laser array for wavelength accuracy and need to verify each channel independently
You are doing silicon photonics device characterization where ring resonator or grating coupler response needs to be measured as a function of wavelength
You need to sweep through 1271/1291/1311/1331 nm in sequence to test all four channels of a 400G-FR4 transceiver on a single bench

A fixed-wavelength 1310 nm source is enough if:

You are doing fiber span loss measurements and just need a clean 1310 nm launch
You are testing return loss (ORL) or back reflection on a link
You are qualifying a receiver's sensitivity at 1310 nm
You are doing power meter calibration or loss budget verification on a single-wavelength system
You are teaching fiber optics or running a training lab where the wavelength itself is not the variable under test

Fixed sources like the HP/Agilent 8167A or a standalone 1310 nm laser diode module are a fraction of the cost and easier to calibrate and maintain. If your application is on that second list, save the budget.

The models that come through our shop

Keysight / Agilent / HP 8168F and 8168C

The 8168F is the most capable O-band tunable laser we regularly see. It covers 1260-1360 nm, tunes in 0.001 nm steps, outputs up to +7 dBm, and has a linewidth specification of less than 100 kHz — tight enough for coherent test applications. The 8168C is an older sibling covering the same band with slightly less output power and a wider linewidth spec. Both use pluggable DFB modules inside an Agilent mainframe.

The most common problem on used 8168F units is degraded output power. The DFB laser chip inside ages and loses output efficiency, usually showing up as the power reading on the front panel being fine while the actual output at the APC connector is 2-3 dB low. We measure actual output with a calibrated 81623B sensor before every shipment. A unit where the power reads right on the display but is short at the connector has a bad output coupler or a degraded isolator — repairable but not cheap.

Wavelength accuracy is the other thing to verify. The 8168F spec is ±0.3 nm absolute, but we have seen units drift to ±0.8 nm after years of storage. A unit with drifted wavelength calibration looks fine on the panel display (it shows the set point) but is actually outputting a different wavelength. If your application requires accuracy better than ±0.5 nm, verify it with a wavemeter before assuming the unit is in cal.

Keysight 8167A (fixed, 1310 nm)

Straightforward fixed-wavelength source, 1310 nm, ±1 nm wavelength accuracy, output typically around 0 dBm. These are workhorses — we find them in fiber test labs that have been running the same loss measurement workflow for twenty years. They fail rarely. When they do fail, it is almost always the output connector being damaged or the internal laser diode reaching end of life. A used 8167A in good condition is one of the most reliable pieces of optical test equipment you can buy.

Anritsu MG9638A and MG9639A

Anritsu's O-band tunable laser covering 1260-1360 nm. The MG9638A has a linewidth below 1 MHz and output up to +6 dBm. The MG9639A is a narrower-linewidth variant (below 100 kHz) aimed at coherent applications. Build quality is good and parts are still available through Anritsu service.

On the used market, the most frequent cosmetic issue is the APC output connector being scratched or chipped from repeated insertions without end-face cleaning. We replace the front panel APC bulkhead before shipping any unit where the connector shows visible end-face damage. A scratched fiber connector tip transfers contamination to every fiber you plug in — not something to ignore. The other thing we check is whether the key-lock is functioning, since stuck key-locks are common on older Anritsu units and can prevent changing settings.

HP/Agilent 8153A with 81533B O-band module

The 8153A is a modular lightwave multimeter mainframe. Pair it with the 81533B O-band laser module and you get a 1310 nm fixed source that also has a power meter port in the same box. Useful for labs that want to do source and measurement without two separate instruments. The 81533B module is a fixed-wavelength DFB at 1310 nm, not tunable. These show up at auction frequently because labs that upgraded to tunable sources no longer needed them.

What we check before shipping any O-band laser

Every O-band source that goes out from us goes through the same checklist. This is not marketing — it is what we actually do because we got burned enough times early on shipping units we had not fully characterized.

Output power at the front connector. Measured with a calibrated 81623B sensor in an 8163B mainframe. Compared against the datasheet spec. Any unit more than 1 dB below spec gets flagged.
Wavelength accuracy. Checked with an HP 86120C multi-wavelength meter for units where the application cares. For fixed-wavelength sources we do a quick OSA check. For tunables we verify calibration at 1310 nm center and at the band edges.
Output connector. Cleaned and inspected under a Noyes FiberChek2 scope. Any end-face that shows scratching in the fiber core area gets a new bulkhead connector.
Stability check. Run the source at full output for 20 minutes and verify power is stable to within 0.1 dB. An unstable source is a bad source regardless of what the datasheet says.
Front panel and controls. Every button, every setting, every displayed value verified functional.

The question we get asked most often

"I need an O-band source for 400G transceiver incoming inspection — what do I actually buy?"

The honest answer depends on how many channels you need to test and how fast. If you are qualifying individual transceivers at 1311 nm only (the center channel), a fixed-wavelength 1310 nm source and a calibrated power meter is sufficient. If you need to sweep all four O-CWDM channels (1271, 1291, 1311, 1331 nm), you need a tunable source with accurate absolute wavelength — a Keysight 8168F in calibration, an Anritsu MG9638A, or a JDSU/VIAVI tunable O-band source.

Production test at volume is different from incoming inspection at low volume. For production, the per-unit test time matters and you want a source with fast settling when you change wavelength — the 8168F settles in under 1 second after a wavelength step, which is fast enough for manual testing but potentially limiting on an automated tester. For incoming inspection where you test maybe ten units a day, settling speed is irrelevant and any of these sources will work.

Frequently asked questions

What wavelength range is O-band and why is 1310 nm the reference center? +

O-band covers 1260 to 1360 nm. The 1310 nm reference point is where single-mode fiber has its zero-dispersion wavelength per ITU-T G.652. A pulse launched at 1310 nm travels without the chromatic dispersion that broadens pulses at other wavelengths, letting you push high data rates through standard SMF without dispersion compensation modules. That property made it the wavelength of choice for early Ethernet over fiber and SONET OC-48, and it is still why 400G-DR4 transceivers use four O-band channels at 1271, 1291, 1311, and 1331 nm.

What is the difference between a tunable O-band laser and a fixed-wavelength O-band laser? +

A tunable laser sweeps across the O-band (typically 1260-1360 nm) and lets you test at any wavelength in that range — essential for characterizing DFB arrays, multiplexers, or CWDM devices where you need to map the response at many wavelengths. A fixed-wavelength laser locks to one wavelength (usually 1310 nm) and is cheaper, more stable over time, and enough for loss budget tests, back-reflection measurements, receiver sensitivity tests, or ORL. For transceiver manufacturing test where you need all four O-CWDM channels, you either use a multi-channel fixed-wavelength module or a tunable that steps through each channel.

How do I verify output power and wavelength accuracy on a used Keysight 8168F? +

Use a calibrated optical power meter (Keysight 8163B with 81623B sensor, or equivalent) to measure output power at the front panel connector. The 8168F is rated +7 dBm typical with ±0.5 dB power accuracy. Wavelength accuracy requires a wavemeter — the HP 86120C or Agilent 86142B spectrum analyzer can resolve it to ±0.001 nm. If the output is more than 1.5 dB below spec and the front panel shows no error, measure through the internal 8168A mainframe power monitor. If the panel display and the measured output disagree, the issue is the output connector or internal coupler, not the laser chip itself.

Can I use an O-band laser source for OTDR calibration? +

Yes, with conditions. An O-band CW laser can calibrate the receiver sensitivity and event detection threshold of a 1310 nm OTDR port if the OTDR accepts an external input. For loss scale calibration — the dB/km accuracy — you need a calibrated launch power and a reference fiber with known loss. The laser source is half of the calibration setup but not the whole thing. If you are calibrating for NIST traceability, the power meter used to quantify launch power needs its own calibration certificate. Skipping that step breaks the traceability chain.

What is the difference between the Keysight 81600B and the 8168F for O-band testing? +

The 81600B covers both C and L band and does not natively cover O-band without the optional 81618A O-band module, which is rare on the used market. The 8168F is dedicated to O-band (1260-1360 nm). If you need O-band specifically, the 8168F or its predecessors are the cleaner choice and easier to source. The 81600B is the right tool when you need the same hardware to cover C-band for coherent testing and occasionally dip into O-band — an uncommon requirement outside of vendor labs building optical amplifier test benches.

Why did demand for O-band laser sources increase so much starting in 2023? +

Two converging trends. First, 400G datacenter interconnect standardized on O-band CWDM channels (400G-DR4, 400G-FR4, 400G-LR4), which pushed hyperscaler optical test setups heavily into O-band for production qualification and incoming inspection. Second, silicon photonics foundries ramping 400G transceiver production need to characterize O-band waveguide loss and DFB array spacing at wafer level. The same sources that sat idle in telecom labs after SONET deployments slowed became critical again in both datacom manufacturing and academic silicon photonics research simultaneously.

Is the Anritsu MG9638A good for O-band testing? +

It is a solid unit. The MG9638A covers 1260-1360 nm with a linewidth below 1 MHz and output power up to +6 dBm. Its step resolution is 0.001 nm, sufficient for CWDM channel testing. The main limitation is that it does not cover C-band — if your test plan also includes 1550 nm you need a separate source. On the used market, the most common issue is the front panel APC connector being scratched or the key lock being stuck. Neither is a laser problem, both are cheap fixes. We verify linewidth with a self-homodyne measurement before shipping any MG9638A.