Detect precision loss

R2026b

Diagnostic action to take if the software detects parameter precision loss

Model Configuration Pane: Diagnostics / Data Validity

Description

The Detect precision loss parameter specifies the diagnostic action to take if the software detects parameter precision loss. Precision loss occurs when the data type of a parameter does not have enough precision to represent the parameter value exactly. As a result, the value of the parameter in the model differs from the ideal value.

Both parameter precision loss and parameter overflow are quantization errors. While parameter precision loss differs from parameter overflow, the distinction can be nuanced. Parameter precision loss occurs when the ideal parameter value is within the range of the data type and scaling, but the data type does not have an exact representation of the parameter value.

The Detect overflow diagnostic reports all quantization errors greater than the specified Bits of error threshold, which has a default setting of one bit. For very small parameter quantization errors and when the Bits of error threshold is set to One bit, precision loss is reported rather than an overflow when:

$$\left(Max+Slope\right)\ge V_{ideal}>\left(Min-Slope\right)$$

where:

• Max is the maximum value the parameter data type can represent.

• Min is the minimum value the parameter data type can represent.

• Slope is the slope of the parameter data type. For integers, the slope is 1.

• V_ideal is the ideal value of the parameter.

Settings

warning (default) | error | none

warning

The software issues a warning if the software detects parameter precision loss.

error

The software issues an error and terminates the simulation if the software detects parameter precision loss.

none

The software does not issue a diagnostic if the software detects precision loss.

Examples

Expected Parameter Precision Loss Warning Not Reported

Troubleshoot when an expected parameter precision loss is not reported by the software.

In this model, the Constant blocks x and y are each set to a value of pi using an unsigned 16-bit integer data type.

With the Detect precision loss parameter set to "warning", the software reports precision loss only for the Constant block y.

Warning: Precision loss occurred for Value (3.1415926535897931) of 'y'. The value cannot be 
represented exactly using the run-time data type 'uint16'. A small quantization error has 
occurred. To disable this warning or error, in the Configuration Parameters > Diagnostics > 
Data Validity pane, set the 'Detect precision loss' option
in the Parameters group to 'none'. 

To understand why the Constant block x does not report a precision loss warning, examine the definitions of these constant values in the base workspace.

x = fi(pi,0,16,0)
y = Simulink.Parameter(pi);
y.DataType = 'uint16'

x = 

     3

          DataTypeMode: Fixed-point: binary point scaling
            Signedness: Unsigned
            WordLength: 16
        FractionLength: 0

y = 

  Parameter with properties:

          Value: 3.141592653589793
     Complexity: 'real'
     Dimensions: [1 1]
      CoderInfo: [1×1 Simulink.CoderInfo]
    Description: ''
       DataType: 'uint16'
            Min: []
            Max: []
           Unit: ''

x and y both specify a value of pi using an unsigned 16-bit integer data type. However, x is specified as a fi object while y is specified as a Simulink.Parameter object.

The fi and Simulink.Parameter object constructors treat the constant differently. The fi constructor quantizes the constant value before storing it. The Simulink.Parameter constructor first stores the constant value using double-precision then casts to the fixed-point data type, causing a precision loss warning to be reported when the model diagram is updated.

Recommended Settings

Application	Setting
Debugging	No impact
Traceability	No impact
Efficiency	No impact
Safety precaution	error

Programmatic Use

Parameter: ParameterPrecisionLossMsg

Type: string | character vector

Value: "none" | "warning" | "error"

Default: "warning"

Version History

Introduced before R2006a