Converting microscopic biophysical perturbations into clean microvolt electrical signals:
Precessing spins induce EMF in conductive copper loop:
\(\mathcal{E} = -\frac{\partial}{\partial t}\int \vec{B}_1^- \cdot \vec{M} dV\)
Principle of Reciprocity
Photons generate electron-hole pairs in SiPM/APD:
\(I_{\text{photo}} = \mathcal{R} \cdot P_{\text{opt}}\)
Internal avalanche gain (\(M \sim 10^6\))
Scalp ionic current converts to wire electron flux:
\(\text{Ag} + \text{Cl}^- \rightleftharpoons \text{AgCl} + e^-\)
Reversible half-cell potential
Acoustic strain waves deform crystal lattice:
\(V = g_{33} \cdot T \cdot t_{\text{crystal}}\)
PZT / CMUT resonance
Every dissipative resistance \(R\) generates spontaneous voltage fluctuations due to thermal agitation of electrons:
At \(310\,\text{K}\) (body temp), a \(50\,\Omega\) resistor yields \(0.93\,\text{nV}/\sqrt{\text{Hz}}\). Over \(25\,\text{kHz}\) bandwidth, this equals \(147\,\text{nV}_{\text{rms}}\).
\(R_{\text{sensor}}\): Ohmic resistance of copper coils, electrodes, or semiconductor bulk.
\(R_{\text{sample}}\): Dissipation from induced eddy currents circulating inside the conductive human brain (\(\sigma \sim 0.4\,\text{S/m}\)).
\(Q_U / Q_L < 1.5\)
Noise originates in coil resistance. Common at low magnetic fields (\(< 0.1\,\text{T}\)) or with sub-millimeter microcoils.
Cooling to 77 K boosts SNR!
\(Q_U / Q_L \ge 2.5 - 5.0\)
Body tissue eddy currents generate \(>80\%\) of total noise. Standard in clinical MRI (\(1.5\,\text{T} - 7\,\text{T}\)).
The patient is the dominant noise source!
When sample-noise dominated, reducing coil wire resistance yields negligible returns.
Instead, optimize filling factor and array channel count.
Every centimeter of coaxial cable between sensor and LNA adds capacitive loading and insertion loss (\(0.1\,\text{dB/m}\)).
Rule of Thumb:
Place preamplifiers directly on the coil/sensor housing. Never route raw microvolt RF signals through long patient cables!
Combining \(N\) decoupled array elements with Roemer optimal weightings yields up to \(2\times - 5\times\) higher SNR in the cerebral cortex than any single volume resonator.
\(\mathbf{\Psi}\) is the noise covariance matrix.
Arrays replace slow gradient phase-encoding steps with spatial sensitivity encoding (SENSE / GRAPPA):
Undersampling k-space by factor \(R\) speeds up image acquisition by \(R\)-fold.
Penalty: \(\sqrt{R}\) loss from fewer sampled data points.
Quantifies noise amplification from ill-conditioned sensitivity profile inversion.
Design Goal: \(g \le 1.15 - 1.30\) across brain cortex.
When two resonant loops sit adjacent to one another, mutual inductance \(M_{12}\) causes severe resonance splitting:
To operate a phased array as independent channels, inter-element isolation must satisfy:
\(S_{21} \le -18\,\text{dB}\) (loaded)
Less than \(1.5\%\) power transfer between adjacent channels.
Adjacent loop coils are overlapped by a critical distance \(d_{\text{overlap}} \approx 0.1 \times \text{diameter}\).
Achieves \(S_{21} < -25\,\text{dB}\) between immediate neighbors.
Limitation:
Only decouples nearest neighbors! Cannot decouple next-nearest neighbors across a curved helmet.
The groundbreaking technique introduced by Roemer et al. (1990) to decouple non-adjacent coils:
The LNA is designed with a very low input impedance:
\(Z_{\text{in}} \approx 1 - 3\,\Omega\)
A \(\lambda/4\) coaxial cable or lumped \(\pi\)-matching network transforms \(Z_{\text{in}}\) into a parallel high impedance:
\(Z_{\text{coil\_open}} \gg 1000\,\Omega\)
Blocks circulating RF current in the loop (\(I_{\text{loop}} \approx 0\)), suppressing secondary magnetic field re-radiation to all other coils!
When multiple sources and detectors illuminate the scalp simultaneously, detectors can confuse light from adjacent optodes.
Solutions:
High skin-electrode contact impedance (\(5 - 20\,\text{k}\Omega\)) converts environmental 60 Hz electric field noise into differential voltage.
Solutions:
| Parameter | Setup | Target Spec | Physical Significance |
|---|---|---|---|
| Return Loss (\(S_{11}\)) | Reflection on loaded coil port | \(< -20\,\text{dB}\) | Impedance matched to \(50\,\Omega\); minimal reflected signal |
| Isolation (\(S_{21}\)) | Transmission between ports | \(< -18\,\text{dB}\) | Channel decoupling; suppression of inductive crosstalk |
| Loaded \(Q\) (\(Q_L\)) | \(-3\,\text{dB}\) bandwidth with phantom | \(f_0 / \Delta f_{-3\text{dB}}\) | Measures total system dissipation (coil + head tissue) |
| \(Q\)-Ratio (\(Q_U / Q_L\)) | Unloaded vs loaded \(Q\) | \(> 2.5 - 4.0\) | Confirms sample-noise dominated operation |
Use our interactive hardware budget engine to configure your team's sensor array:
Calculate thermal Johnson noise \(V_n\), spectral density (\(\text{nV}/\sqrt{\text{Hz}}\)), and check your \(Q_U / Q_L\) noise regime.
Formulate critical overlap, preamplifier decoupling, or active DRL shielding to hit \(S_{21} < -18\,\text{dB}\).
Simulate depth roll-off ($1/r^2$ or optical diffusion) and estimate parallel imaging $g$-factor bounds.
Magn. Reson. Med., 16(2): 192–225
The landmark paper establishing NMR phased arrays, preamplifier decoupling, and SNR combination.
J. Magn. Reson., 24(1): 71–85
The foundational derivation of the NMR reciprocity theorem and fundamental thermal noise limits.
Magn. Reson. Med., 42(5): 952–962
SENSE parallel imaging, mathematical unfolding of sensitivity fields, and geometry factor derivation.
Next: Optical Imaging & In-Class fNIRS Demonstration