“What size are my particles, right now, in the reactor? Has a second population appeared? Has agglomeration started?” Three questions asked about a concentrated suspension, to be settled without diluting it.
SR-DLS, spatially resolved dynamic light scattering, answers on the medium as it is, inside the reactor. Laboratory particle sizing remains the reference.
Where you meet it: wet milling, silica nanoparticles, injectable nanoemulsions.
The essentials in four points
- Returns a size distribution, from the nanometre to the micron: width, tail towards the coarse end, appearance of a second population.
- In a concentrated, turbid medium, with no dilution: multiple scattering is not corrected, it is spatially excluded.
- No multivariate model: size comes straight out of the physics of the measurement.
- Reference texts: none for the spatially resolved variant; USP <430> in pharma and ISO 22412 elsewhere cover classical scattering, on a diluted sample.
How the measurement works
Size is derived from the Brownian motion of the particles. A low-coherence interferometer slices the medium into depth layers and uses only those still in single scattering: that is what allows measurement without dilution. The chain is detailed in the fold “The four steps of the chain”.
Typical applications
Particle size
Distribution during the process, on the concentrated medium, with no sampling and no dilution.
Agglomeration and crystallisation
Onset of agglomeration, a second population appearing, the end point of a crystallisation.
Application examples
Production of lipid nanoparticles for mRNA vaccines: size controlled in real time.
The measurement is taken through a clean transparent wall, or on a bypass loop — often the more robust configuration, because it gives control of the flow.
Identity card of the technique
| Criterion | SR-DLS |
|---|---|
| What the measurement sees | A size distribution, from the nanometre to the micron, deduced from Brownian motion. A low-coherence interferometer slices the medium by depth and uses only the slices still in single-scattering regime. |
| Selectivity | Not applicable: the quantity is physical. The measurement does not say what the particles are made of. |
| What interferes | Temperature first: size is deduced from a velocity, which depends on viscosity, which depends strongly on temperature. Then bubbles, too broad a distribution, a flow that has left the laminar regime. |
| Sample presentation | Without dilution, in concentrated and turbid media, through a clean transparent wall or on a bypass loop. |
| What the model requires | No model to calibrate or revalidate: size comes out of the signal processing. The effort moves to integration, and to the in-flow protocol, which remains to be built. |
| Reference texts | In pharma: no chapter is devoted to the spatially resolved variant. Dynamic light scattering itself is covered: USP <430>, harmonised and official since 1 May 2024. Outside pharma: ISO 22412, in its 2025 edition. In both cases, attaching the method is examined case by case: these texts describe a measurement in a cell on a diluted sample, which this technique achieves otherwise. |
Spatial exclusion: the point that decides everything
Here is the point that carries everything. The method sets multiple scattering aside spatially rather than correcting it. The depth slices in the multiple-scattering regime receive zero weight. The slices where the light was scattered once carry the result.
That is the entire foundation of measurement with no dilution. A signal that would mislead is left out rather than straightened. It also explains why the method calls for no multivariate model: the size comes out of the physics rather than out of a regression on reference values. What the measurement will ask you to supply.
What the measurement returns
A distribution, from the nanometre to the micron
A distribution rather than a single value. Reducing it to its mean usually loses the useful information: the width, the tail towards the large, a second population appearing.
In a concentrated, turbid medium, undiluted
The direct consequence of the spatial exclusion. The medium can stay opaque to the eye. What counts is that somewhere in depth a slice remains in the single-scattering regime.
Through a wall, or on a bypass
Through a clean transparent wall, or on a bypass loop. The loop is most often the sounder configuration, because it gives control of the flow.
The four steps of the chain
- A weakly coherent source. Its coherence length, of the order of 4 µm, sets how finely a signal can be assigned to a given depth. That is the depth resolution, and it is a property of the source rather than a setting.
- A fast spectral acquisition. The interference spectra are recorded at a rate reaching 76 kHz. That rate is the rate of the raw signal rather than of the results. The temporal fluctuations are derived from it afterwards.
- An inverse Fourier transform. It converts each spectrum into a depth profile, and allows up to 1,024 correlation functions to be computed at once, one per depth slice of the medium.
- An inversion towards a size distribution. Each correlation function translates the speed of the Brownian motion at that depth, from which a hydrodynamic diameter follows through the Stokes-Einstein relation.
The qualification chain: what carries figures, and what to build
On this point the available documentation is strong. The factory acceptance protocols and then the installation and operational qualification carry numerical acceptance criteria, which is far from always the case.
- an accuracy of ± 5 % on a control material.
- a relative standard deviation under 5 % in repeatability.
- a stability verified at three standard deviations over 30 successive measurement points.
- a hysteresis check on the motorised parts.
- an endurance test of 8 to 10 hours.
That is a real working base. It leaves two things to build, and they belong here, because an auditor will look for them.
The traceability chain runs to the control material
The protocols rest on a calibrated reference suspension, at 100 nm, with a numerical acceptance criterion. Beyond that, the chain linking the returned value to a national reference is still to be documented: no certified reference material producer, no national metrology institute, no calibration accreditation. That leaves the accuracy to be established at your site, by certified standards or an independent method. It is analytical procedure validation work, costed at scoping rather than discovered at inspection.
Performance in flow is the protocol to build
This is the most interesting gap in the field, and the most useful to know. The qualification covers the instrument at rest. The promise covers the measurement in motion. And flow is precisely what you come in-line to measure. This is not our observation: Ph. Eur. 5.25 states that the qualification criteria of pharmacopoeial techniques were designed for off-line analytical systems, and are not always relevant or practical in a PAT setting. The answer is to build what is missing: a verification protocol under real conditions, with a control material, a defined flow range and a periodicity. It is not supplied. It is designed.
Integration with the control system, and what to settle
This is a real strength, and a rare one. A standardised industrial communication interface exposes the complete size distribution and the velocity profile as tables rather than a few summary indicators alone. Alongside them sit around sixty parameters accessible for writing, so control rather than a simple read. And fifteen analysis quality criteria with a configurable binary filter, which sets a doubtful measurement aside automatically before it enters a decision.
The most interesting point for an integrator sits elsewhere: temperature and viscosity overrides are accessible for writing. A third-party sensor can therefore correct the Stokes-Einstein relation in real time, which is to say compensate the main source of systematic error in a Brownian motion measurement, with a value measured in the right place rather than an assumed one.
Three points to settle before the architecture
No standardised information model, no documented namespace: the integration is built on a structure specific to the manufacturer, which carries a maintenance cost and moves at each major version. No security policy is published. And the communication server stays active after the main software is closed and accepts writes from third-party clients. A characteristic to handle attentively in a regulated environment, where access control and data integrity are requirements. See what "21 CFR Part 11 ready" means, and what it leaves with you.
An example of a closed loop, published by a manufacturer in an application note (2025): on a continuous liposome line using turbulent jet injection, the mean size measured in-line sets the aqueous phase flow rate through a PID controller. The 85 nm set point is reached in under ten minutes, while the flow rate goes from about 130 to 100 mL/min, then held continuously. The standard deviation of the mean size is 3 nm with control, against 4 nm without. It is manufacturer data, from one trial: it shows the loop can be built; its robustness is verified on your process.
The bounds, as the manufacturer writes them
They are published here unretouched, and deliberately so: documentation that writes its own bounds is what lets you decide before the trial. They fall into three families.
| Published bound | Family | What it calls for |
|---|---|---|
| Laminar flow required | Hydraulic | The design of the loop is the heart of the project, ahead of the choice of sensor |
| Air bubbles are governing | Hydraulic | Degassing, pipework routing and tapping point to be studied |
| Vibration to be kept out | Mechanical | A mount decoupled from the machine |
| Shear above the diffusion rate degrades the signal | Hydraulic | The flow rate is set on the target size rather than on circulation comfort |
| Two populations whose size ratio is under 3 are reported together | Physics of the method | Where the question is a close bimodality, another measurement answers it |
| A polydispersity index above 0.3 | Physics of the method | The default interpretation no longer holds: the result calls for further analysis, failing which it reads as a trend. A field rule, not a normative one — USP <430> defines the index without setting a limit on it |
| Size, and shape or zeta potential elsewhere | Domain of use | Agglomeration and growth give the same evolution: a second measurement separates them |
| Explosive-atmosphere certification is a made-to-order configuration | Domain of use | Compliant configurations exist, made to order: a question to put first where your area calls for it, rather than last |
| Range from 0 to 70 °C | Domain of use | A hotter process calls for an exchanger on the bypass, with the transit time that comes with it |
The first family is pipework rather than optics. That is where the real feasibility of most projects is decided, far more than in the specifications of the instrument.
Two different results, one convention apart
A matter of calculation basis, and it carries figures
On the same injectable lipid emulsion, two dynamic light scattering devices return 332 nm and 369 nm — two Zav values, the intensity-weighted mean diameter — about 11 % apart. Amply enough to open an investigation. Converted to a volume distribution, both results become 415 ± 5 nm.
The gap was not an instrument error. Both devices return the same quantity, but they neither illuminate at the same wavelength nor collect at the same angle — 173° against 180° — and at this size we have left the Rayleigh regime: the weight each size carries in the signal becomes specific to the instrument. The volume distribution depends on neither of those two choices, which is why both devices recover the same value there. The consequence is documentary rather than technical. The calculation basis, the detection angle and the inversion algorithm belong in the specification, alongside the target value. The problem is strictly the same as transferring a model between two spectrometers.
Always look at what the measurement was compared against
In the application literature of this family, the reference method is very often the same device in another configuration. That demonstrates reproducibility rather than accuracy. Two things establish accuracy: certified standards, or an independent method whose calculation basis is stated. So the question to put to a supplier, and to us, stays the same: what did you compare against, and with what traceability?
What is worth preparing on your side
- The target nanometric range, and above all the real concentration of the medium. It is the second that decides whether SR-DLS applies, more than the first.
- The hydraulic conditions really available: flow rate, regime, bubbles, viscosity, temperature at the measurement point. That is the heart of the subject.
- The current reference method and its calculation basis. Without it the trial is judged on a gap that is not one.
- The area and temperature requirements, set against the bounds in the table above. Two lines are often enough to conclude.
- An automation contact involved from the start. An integration that exposes parameters for writing is handled early rather than at the end of a project.
Frequently asked questions
How can the measurement work in an opaque medium?
Because it sets out to read a slice rather than to see through. It cuts the medium into depth slices and uses only those where the light was scattered once, typically the ones nearest the window. A very concentrated medium thins that useful zone. That is how concentration ends up bounding the method, rather than through a displayed threshold.
Is there a model to build and to maintain?
For the size itself, no: it comes from the signal processing, out of the physics of Brownian motion. No reference sample campaign, no model to revalidate. The effort moves elsewhere, and it is real: to the physical integration, flow, bubbles, vibration, temperature, and to the demonstration of accuracy, which is established on your product.
Why does temperature keep coming back?
Because the size is derived from a Brownian motion speed, which follows the viscosity of the medium, which follows the temperature strongly. A temperature error translates directly into a size error. That is where the overrides accessible for writing earn their keep. A probe placed in the right spot corrects the calculation in real time, rather than leaving the instrument to work on an assumed value.
What measurement rate can be expected?
It follows the configuration entirely: concentration, size range, integration time retained, loop flow rate. The frequencies quoted in the description of the arrangement are raw signal acquisition frequencies rather than rates of usable results. Confusing the two leads to promises that cannot hold. The sound approach is to fix the rate your process needs first, then verify that it is reachable in your conditions.
Can it measure directly in the reactor?
Through a clean transparent wall, in certain configurations, yes. Two points often decide. Fouling on the product side, and the fact that the layer measured runs along the wall, so it represents the volume where the flow brings the volume to it. A bypass loop stays the sounder answer in most cases, at the price of a transit time to cost.
How do we know whether it applies to us, without committing to a project?
Three pieces of information usually settle it: the size range, the expected polydispersity index, and the flow regime available. Where all three pass, a proof of concept on your real medium is the logical next step. Where one of the three points elsewhere, knowing it straight away is a useful answer, and it costs nothing.
Describe your medium and your flow conditions. That is what decides, well ahead of the sensor.
Forty-five minutes is enough: whether measurement with no dilution applies to your suspension, which hydraulic configuration it supposes, and what would have to be built to make it defensible at inspection. Where the distribution is very wide, the bubbles unavoidable or the area incompatible, we will say so — and where to look instead.