1. Introduction: Two Parameters, One System Trade-off

Every coherent DWDM channel occupies a defined slice of the optical spectrum. Two numbers govern what happens inside that slice: the baud rate — how many symbols per second the transmitter launches into the fiber — and the channel width — how many GHz the wavelength-selective switch (WSS) opens to let that signal pass. These two parameters are not independent. The baud rate determines the minimum spectral width the signal physically requires. The channel width determines the maximum spectral window the photonic layer will provide. The gap between them — or the lack of one — controls spectral efficiency, filter-induced penalty, inter-channel crosstalk, and ultimately the total capacity a fiber can carry.

At 35 GBaud, a dual-polarisation QPSK signal carrying 100 Gb/s occupies approximately 37.5 GHz of spectrum with practical pulse shaping, fitting comfortably inside a 50 GHz fixed-grid slot with ~12.5 GHz of guard space on each side. At 95 GBaud, a shaped 400 Gb/s signal needs approximately 100 GHz of spectral occupancy and demands either a 112.5 GHz or 150 GHz flex-grid slot depending on the guard-band and filter-edge requirements. At 200 GBaud, a single-carrier 1.6 Tb/s signal requires approximately 200 GHz of channel width, and the entire C-band holds only 24 such channels. Each of these configurations represents a fundamentally different engineering trade-off between per-channel capacity, total fiber capacity, and transmission reach.

This article dissects that trade-off with quantitative precision. It covers the physics linking baud rate to occupied bandwidth, the standards framework that governs channel allocation, the impact of modulation format selection on the channel budget, the penalties imposed by cascaded optical filtering, and the practical design decisions operators face when planning networks from metro rings to transoceanic cables. The analysis draws on current-generation coherent platform capabilities, including systems operating at baud rates from 60 GBaud to 200 GBaud across both pluggable and embedded form factors.

Engineering context: The channel-width-to-baud-rate ratio is the single most underappreciated parameter in DWDM link engineering. An operator who packs 64 GBaud signals into 50 GHz channels will experience 3–5 dB of filter-induced penalty on the first ROADM pass. An operator who allocates 100 GHz to a 35 GBaud signal wastes 62.5 GHz of spectrum per channel — enough to carry an additional 100G wavelength. Both choices cost real money. Understanding the interplay prevents both.

2. Foundational Physics: Baud Rate, Spectral Width, and the Nyquist Limit

2.1 Baud Rate: The Clock of the Optical Channel

Baud rate (symbol rate, Rs) measures the number of discrete symbol transitions per second on the optical carrier. Each symbol carries log2(M) bits, where M is the modulation order — 2 for BPSK, 4 for QPSK, 16 for 16QAM, 64 for 64QAM. Dual-polarisation coherent detection doubles the information capacity by transmitting independent symbol streams on the X and Y polarisation states of the same wavelength. The gross data rate is therefore:

Rgross = Rs × log2(M) × Npol

Where:
  Rs     = Symbol rate (GBaud)
  M      = Modulation order (4 = QPSK, 16 = 16QAM, 64 = 64QAM)
  Npol  = Number of polarisations (2 for DP coherent)

Practical Example — 800G DP-16QAM:
  Rgross = 118 GBaud × 4 bits/sym × 2 pol = 944 Gb/s
  Net rate after ~15% SD-FEC overhead: ~800 Gb/s

The baud rate is the clock rate of the optical modulator and the analogue-to-digital converters inside the coherent DSP. Increasing it demands wider-bandwidth electro-optic components — modulators, drivers, trans-impedance amplifiers, and ADCs/DACs that can handle the analogue waveform without distortion. A 200 GBaud design requires approximately 100 GHz of electrical baseband bandwidth from the front-end optics and converters, compared to approximately 35 GHz for a 60 GBaud design. That 3× bandwidth increase is the primary reason why each baud-rate generation requires a new silicon process node and new photonic integration. For a deep exploration of how modulation format selection interacts with baud rate to determine bit rate, spectral width, and OSNR requirements, see the MapYourTech reference on Bit Rate vs Baud Rate in Optical Networks.

2.2 The Nyquist Limit and Occupied Bandwidth

The Nyquist-Shannon sampling theorem establishes that the minimum bandwidth required to transmit Rs symbols per second without inter-symbol interference (ISI) is Rs/2 Hz on each side of the carrier — a total one-sided bandwidth of Rs/2. In a dual-sideband system with no excess bandwidth, the minimum occupied bandwidth equals the baud rate itself: Bmin = Rs. This is the Nyquist limit, and it represents the theoretical minimum spectral footprint of a coherent signal.

Real transmitters cannot achieve this limit. Practical pulse-shaping filters introduce excess bandwidth characterised by the roll-off factor (α), which ranges from 0 (ideal rectangular spectrum, infinite time-domain sinc pulse) to 1 (fully excess bandwidth, smooth cosine pulse). The occupied bandwidth with a raised-cosine pulse shape is:

BWsignal = Rs × (1 + α)

Where:
  α = Roll-off factor (typically 0.05 to 0.20 for modern coherent)

Practical Examples:
  35 GBaud, α=0.07:   BW = 35 × 1.07 = 37.5 GHz
  64 GBaud, α=0.10:   BW = 64 × 1.10 = 70.4 GHz
  95 GBaud, α=0.06:   BW = 95 × 1.06 = 100.7 GHz
  131 GBaud, α=0.05:  BW = 131 × 1.05 = 137.6 GHz
  200 GBaud, α=0.05:  BW = 200 × 1.05 = 210 GHz

Modern coherent systems use roll-off factors between 0.01 and 0.20. Tighter roll-off (lower α) compresses the signal spectrum and improves spectral efficiency, but it demands more precise filtering in both the transmitter DSP and the line-side optical path, and it increases sensitivity to ISI from imperfect filter responses. A detailed examination of spectral width physics, measurement methods, and practical implications is available in the MapYourTech guide on Spectral Width in Optical Communication.

Figure 1: Occupied signal bandwidth versus roll-off factor for representative baud rates. Every 0.05 reduction in α saves approximately 5×Rs × 0.05 GHz of spectrum — at 200 GBaud, each 0.05 reduction saves 10 GHz of channel width, potentially fitting one additional channel per 120 GHz of band.

2.3 The 3 dB and 10 dB Bandwidth Measures

Coherent transceiver specifications commonly report two bandwidth measures beyond the occupied spectral width. The 3 dB signal bandwidth defines the frequency span at which the signal power spectral density drops 3 dB below its peak. The 10 dB bandwidth defines the span at which it drops 10 dB. These matter because WSS filter pass-bands must be wider than the 10 dB signal bandwidth to avoid clipping spectral content and inducing ISI penalty.

For a 56 GBaud transmitter, the 3 dB bandwidth is approximately 57.7 GHz and the 10 dB bandwidth is approximately 59.9 GHz. For a 35 GBaud transmitter, the corresponding values are approximately 35.5 GHz (3 dB) and 36.8–37.4 GHz (10 dB), with the 250 Gb/s mode showing a slightly wider 10 dB bandwidth of 37.4 GHz due to its specific pulse-shaping profile. These measurements directly inform the minimum WSS filter width: the media channel (MC) must be wide enough to pass the 10 dB bandwidth without truncation, plus the dead-band margins required for WSS pixel roll-off.

Takeaway: The baud rate sets the fundamental spectral footprint of the coherent signal through the relationship BW = Rs × (1 + α). The WSS filter width must exceed this footprint plus guard margins. When these two values converge too closely, filter-narrowing penalty degrades the signal; when they diverge too widely, spectrum goes unused.

3. Channel Width Standards: From Fixed ITU Grid to Flexible Grid

3.1 Fixed Grid: The 50 GHz and 100 GHz Era

ITU-T Recommendation G.694.1 defines the DWDM frequency grid anchored at 193.1 THz (~1552.52 nm). The fixed grid provides uniform channel spacings of 12.5, 25, 50, and 100 GHz. Most deployed DWDM networks use the 50 GHz spacing, which yields approximately 96 channels across the conventional C-band (191.3–196.1 THz). Each channel occupies a fixed 50 GHz filter window defined by the WSS hardware, regardless of the actual signal bandwidth inside it.

Fixed grid works well when all channels use similar baud rates and modulation formats. A 28–35 GBaud DP-QPSK signal carrying 100 Gb/s occupies ~37.5 GHz and fits comfortably in a 50 GHz slot with adequate guard space. The problem emerges when operators want to carry 200G, 400G, or 800G on the same infrastructure. A 64 GBaud DP-16QAM signal at 400 Gb/s needs ~70 GHz of spectral width — 20 GHz wider than a 50 GHz slot allows. Forcing this signal into a 50 GHz filter truncates the spectral tails, induces ISI, degrades OSNR sensitivity by 3–5 dB, and may prevent the DSP from acquiring the signal at all on long-haul routes.

3.2 Flexible Grid: Variable-Width Spectral Slots

The 2012 revision of G.694.1 introduced the flexible grid (commonly called flex-grid or flexi-grid), which decouples channel allocation from a fixed spacing. The flex-grid defines two independent granularity parameters: a 6.25 GHz central-frequency granularity and a 12.5 GHz slot-width granularity. Any channel can be assigned a spectral slot whose width is an integer multiple of 12.5 GHz (m × 12.5 GHz, where m ≥ 1), centred on a frequency that aligns to the 6.25 GHz grid.

This produces channel widths of 12.5, 25, 37.5, 50, 62.5, 75, 87.5, 100, 112.5, 125, 137.5, 150 GHz, and so on. The channel width is selected to match the actual spectral occupancy of each signal, eliminating both spectral waste and filter squeeze. A 100G signal at 35 GBaud can be allocated a 50 GHz slot (m = 4). A 400G signal at 64 GBaud can use a 75 GHz slot (m = 6). An 800G super-channel built from two 64 GBaud sub-carriers can be allocated a single 150 GHz slot (m = 12) with the sub-carriers packed tightly inside. For a complete treatment of the G.694.1 framework including fixed and flexible grid mechanics, frequency calculation, and flex-grid slot allocation tables, see the MapYourTech guide on ITU-T G.694.1 DWDM Channel Grid.

Premium Article — Free 20% Preview

Read the Full Analysis with Premium

The remaining 80% of this article — the design numbers, trade-offs and field guidance — is part of MapYourTech Premium, along with the full premium library, courses and professional tools.

Instant access · Cancel anytime · 48-hour trial available
Sanjay Yadav

Optical Communications & Network Automation Expert | Author of 3 Books for Optical Engineers | Founder, MapYourTech

Optical networking engineer with nearly two decades of experience across DWDM, OTN, coherent optics, submarine systems, and cloud infrastructure. Founder of MapYourTech.

Follow on LinkedIn

Related Articles on MapYourTech